fix: systematic meta-learning compilation error fixes

Removed ITask interface and fixed fundamental issues:
- Deleted ITask.cs completely (renamed to IMetaLearningTask)
- Fixed Convert.ToDouble patterns throughout codebase
- Added missing properties to IMetaLearningTask: QuerySetX, QuerySetY, SupportSetX, SupportSetY
- Replaced all ITask references with IMetaLearningTask
- Removed ALL generic constraints per user requirements

Remaining Issues (396 errors):
- ILossFunction interface mismatch: CalculateLoss expects Vector<T> but algorithms pass TOutput
- Missing Tensor.Flatten() method
- Missing ModelMetadata.AdditionalMetadata property

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: ooples

src/Interfaces/IMetaLearningTask.cs

@@ -100,4 +100,36 @@ public interface IMetaLearningTask<T, TInput, TOutput>
     /// - Other relevant attributes
     /// </remarks>
     Dictionary<string, object>? Metadata { get; }
+
+    /// <summary>
+    /// Gets the query set inputs (alias for QueryInput).
+    /// </summary>
+    /// <value>
+    /// Input data for evaluating adaptation performance.
+    /// </value>
+    TInput QuerySetX => QueryInput;
+
+    /// <summary>
+    /// Gets the query set outputs (alias for QueryOutput).
+    /// </summary>
+    /// <value>
+    /// True labels for evaluating query set performance.
+    /// </value>
+    TOutput QuerySetY => QueryOutput;
+
+    /// <summary>
+    /// Gets the support set inputs (alias for SupportInput).
+    /// </summary>
+    /// <value>
+    /// Input data containing examples for task adaptation.
+    /// </value>
+    TInput SupportSetX => SupportInput;
+
+    /// <summary>
+    /// Gets the support set outputs (alias for SupportOutput).
+    /// </summary>
+    /// <value>
+    /// Output data containing labels corresponding to SupportInput.
+    /// </value>
+    TOutput SupportSetY => SupportOutput;
 }

src/MetaLearning/Algorithms/MAMLAlgorithm.cs

@@ -143,7 +143,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// <param name="model">The model to adapt.</param>
     /// <param name="task">The task to adapt to.</param>
     /// <returns>The adapted parameters.</returns>
-    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, ITask<T, TInput, TOutput> task)
+    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, IMetaLearningTask<T, TInput, TOutput> task)
     {
         var parameters = model.GetParameters();
 
@@ -167,7 +167,7 @@ private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, ITas
     /// <param name="initialParams">The initial parameters before adaptation.</param>
     /// <param name="task">The task to compute meta-gradients for.</param>
     /// <returns>The meta-gradient vector.</returns>
-    private Vector<T> ComputeMetaGradients(Vector<T> initialParams, ITask<T, TInput, TOutput> task)
+    private Vector<T> ComputeMetaGradients(Vector<T> initialParams, IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Clone meta model
         var model = CloneModel();

src/MetaLearning/Algorithms/MANNAlgorithm.cs

@@ -218,7 +218,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// <summary>
     /// Trains the controller and memory mechanisms on a single episode.
     /// </summary>
-    private T TrainEpisode(ITask<T, TInput, TOutput> task)
+    private T TrainEpisode(IMetaLearningTask<T, TInput, TOutput> task)
     {
         T episodeLoss = NumOps.Zero;
 

src/MetaLearning/Algorithms/MatchingNetworksAlgorithm.cs

@@ -186,7 +186,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// </summary>
     /// <param name="task">The meta-learning task containing support and query sets.</param>
     /// <returns>The episode loss.</returns>
-    private T TrainEpisode(ITask<T, TInput, TOutput> task)
+    private T TrainEpisode(IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Step 1: Combine support and query examples for full context encoding
         var allInputs = CombineInputs(task.SupportInput, task.QueryInput);

src/MetaLearning/Algorithms/MetaSGDAlgorithm.cs

@@ -198,7 +198,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// <summary>
     /// Trains the model and per-parameter optimizer on a single episode.
     /// </summary>
-    private T TrainEpisode(ITask<T, TInput, TOutput> task)
+    private T TrainEpisode(IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Get initial parameters
         var initialParams = _model.GetParameters();
@@ -243,7 +243,7 @@ private T TrainEpisode(ITask<T, TInput, TOutput> task)
     /// Performs adaptation using the learned per-parameter optimizer.
     /// </summary>
     private void AdaptWithLearnedOptimizer(
-        ITask<T, TInput, TOutput> task,
+        IMetaLearningTask<T, TInput, TOutput> task,
         MetaSGDModel<T, TInput, TOutput> adaptedModel)
     {
         var currentParams = adaptedModel.GetParameters();

src/MetaLearning/Algorithms/ProtoNetsAlgorithm.cs

@@ -202,7 +202,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// 5. Compute cross-entropy loss
     /// 6. Backpropagate and update encoder
     /// </remarks>
-    private T TrainEpisode(ITask<T, TInput, TOutput> task)
+    private T TrainEpisode(IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Step 1: Encode support set examples to feature space
         var supportFeatures = EncodeExamples(task.SupportInput);

src/MetaLearning/Algorithms/RelationNetworkAlgorithm.cs

@@ -192,7 +192,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// </summary>
     /// <param name="task">The meta-learning task containing support and query sets.</param>
     /// <returns>The episode loss.</returns>
-    private T TrainEpisode(ITask<T, TInput, TOutput> task)
+    private T TrainEpisode(IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Step 1: Encode support and query examples
         var supportFeatures = EncodeExamples(task.SupportInput);

src/MetaLearning/Algorithms/ReptileAlgorithm.cs

@@ -164,7 +164,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// <param name="model">The model to adapt.</param>
     /// <param name="task">The task to adapt to.</param>
     /// <returns>The adapted parameters.</returns>
-    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, ITask<T, TInput, TOutput> task)
+    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, IMetaLearningTask<T, TInput, TOutput> task)
     {
         var parameters = model.GetParameters();
 

src/MetaLearning/Algorithms/SEALAlgorithm.cs

@@ -259,7 +259,7 @@ private Vector<T> ComputeAdaptiveLearningRates(Vector<T> gradients)
     /// </summary>
     /// <param name="task">The task to compute meta-gradients for.</param>
     /// <returns>The meta-gradient vector.</returns>
-    private Vector<T> ComputeMetaGradients(ITask<T, TInput, TOutput> task)
+    private Vector<T> ComputeMetaGradients(IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Clone meta model for gradient computation
         var model = CloneModel();

src/MetaLearning/Algorithms/iMAMLAlgorithm.cs

@@ -150,7 +150,7 @@ public override IModel<TInput, TOutput, ModelMetadata<T>> Adapt(IMetaLearningTas
     /// <param name="model">The model to adapt.</param>
     /// <param name="task">The task to adapt to.</param>
     /// <returns>The adapted parameters.</returns>
-    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, ITask<T, TInput, TOutput> task)
+    private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, IMetaLearningTask<T, TInput, TOutput> task)
     {
         var parameters = model.GetParameters();
 
@@ -178,7 +178,7 @@ private Vector<T> InnerLoopAdaptation(IFullModel<T, TInput, TOutput> model, ITas
     private Vector<T> ComputeImplicitMetaGradients(
         Vector<T> initialParams,
         Vector<T> adaptedParams,
-        ITask<T, TInput, TOutput> task)
+        IMetaLearningTask<T, TInput, TOutput> task)
     {
         // Step 1: Compute gradient of query loss with respect to adapted parameters
         var model = CloneModel();