Create a MemoryPlanningAlgo class. (#11824)

Summary:

Refactor our memory planning algos into a `MemoryPlanningAlgo` base class + algo-specific implementation in derived class. This refactor separates common utility functions + constraint handling to the base class.

This way, we can add support for hierarchical graphs (using maps), and add more types of constraints (like pinning a tensor to specific dtcm bank) without changing the algo itself.

Reviewed By: zonglinpeng

Differential Revision: D76954785

Author: hsharma35

backends/cadence/aot/TARGETS

@@ -448,6 +448,22 @@ python_unittest(
     ],
 )
 
+python_library(
+    name = "memory_planning_algo",
+    srcs = [
+        "memory_planning_algo.py",
+    ],
+    deps = [
+        ":memory_constraints",
+        ":pass_utils",
+        "//executorch/exir:lib",
+        "//executorch/exir:memory_planning",
+        "//executorch/exir:tensor",
+        "//executorch/exir/passes:lib",
+        "fbsource//third-party/pypi/tabulate:tabulate",
+    ],
+)
+
 python_library(
     name = "memory_planning",
     srcs = [
@@ -456,6 +472,7 @@ python_library(
     deps = [
         "fbsource//third-party/pypi/tabulate:tabulate",
         ":memory_constraints",
+        ":memory_planning_algo",
         ":pass_utils",
         "//caffe2:torch",
         "//executorch/exir:lib",