feat: add AWS Bedrock support for codebase indexing

[roomote]

Summary

This PR adds AWS Bedrock support for codebase indexing, addressing issue #8658.

Motivation

Users who utilize AWS Bedrock for their AI workloads previously needed to set up an OpenAI-compatible API gateway (like https://github.com/aws-samples/bedrock-access-gateway) to use codebase indexing. This PR removes that requirement by adding native AWS Bedrock support.

Changes Made

Core Implementation

• ✅ Added bedrock as a new EmbedderProvider type
• ✅ Implemented BedrockEmbedder class with full IEmbedder interface compliance
• ✅ Support for multiple embedding models:
  • Amazon Titan Embed Text v1 & v2
  • Amazon Titan Embed Image v1
  • Cohere Embed English/Multilingual v3

Configuration & Integration

• ✅ Updated configuration manager to handle Bedrock settings (region, profile)
• ✅ Modified service factory to instantiate BedrockEmbedder
• ✅ Added model dimension profiles for all supported Bedrock models

Testing & Quality

• ✅ Comprehensive test suite with 23 test cases
• ✅ Full test coverage including error handling, retries, and edge cases
• ✅ Added localization strings for error messages

Security & Best Practices

• ✅ Uses AWS SDK credential providers (environment variables or profiles)
• ✅ No hardcoded credentials
• ✅ Implements retry logic with exponential backoff for rate limiting
• ✅ Proper error handling with telemetry integration

How to Test

1. Configure AWS credentials (via environment variables or AWS profile)
2. Set up Bedrock configuration in VSCode settings:
   • Set embedder provider to bedrock
   • Configure AWS region (defaults to us-east-1)
   • Optionally set AWS profile name
3. Test codebase indexing functionality

Breaking Changes

None - this is a purely additive change.

Checklist

• Tests pass
• Code follows existing patterns
• Documentation updated (localization strings)
• No security vulnerabilities introduced

Closes #8658

---

Important

Adds AWS Bedrock support for codebase indexing with new embedder provider, configurations, and tests.

• Behavior:
  • Adds bedrock as a new EmbedderProvider type in interfaces/manager.ts and embeddingModels.ts.
  • Implements BedrockEmbedder class in bedrock.ts with full IEmbedder interface compliance.
  • Supports multiple embedding models: Amazon Titan Embed Text v1 & v2, Amazon Titan Embed Image v1, Cohere Embed English/Multilingual v3.
• Configuration & Integration:
  • Updates config-manager.ts to handle Bedrock settings (region, profile).
  • Modifies service-factory.ts to instantiate BedrockEmbedder.
  • Adds model dimension profiles for all supported Bedrock models in embeddingModels.ts.
• Testing & Quality:
  • Adds comprehensive test suite in __tests__/bedrock.spec.ts with 23 test cases.
  • Full test coverage including error handling, retries, and edge cases.
  • Adds localization strings for error messages in embeddings.json.
• Security & Best Practices:
  • Uses AWS SDK credential providers (environment variables or profiles).
  • Implements retry logic with exponential backoff for rate limiting.
  • Proper error handling with telemetry integration.

^{This description was created by for 013496e. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

Consider defining a proper type for the config object instead of using (codebaseIndexConfig as any). This can help avoid runtime errors and improve type safety.

Suggested change

	const openAiCompatibleBaseUrl = (codebaseIndexConfig as any).codebaseIndexOpenAiCompatibleBaseUrl ?? ""
	const openAiCompatibleBaseUrl = codebaseIndexConfig.codebaseIndexOpenAiCompatibleBaseUrl ?? ""

[roomote]

Token limits for Cohere models through Bedrock are incorrectly enforced. Cohere Embed v3 models have a maximum input token limit of 512, but this code uses OpenAI's MAX_ITEM_TOKENS (8191) for all Bedrock models. This will cause API errors when processing texts between 512-8191 tokens with Cohere models, as the batching logic won't filter them out.

Consider adding model-specific token limits similar to how GEMINI_MAX_ITEM_TOKENS is defined, then checking the model ID to apply the appropriate limit. For example:

const maxTokens = model.startsWith('cohere.embed') ? 512 : 8191;
if (itemTokens > maxTokens) { ... }