feature/extend AWS Bedrock node with full model catalog and custom model support

[Ankit5467]

Expand the awsChatBedrock node from limited Claude support to all 63 active
Bedrock models across 16 AWS regions, with automatic inference profile routing.
Add support for imported, fine-tuned, and provisioned-throughput models via a
single "Custom Model ARN" field — the node auto-detects the model type and
routes to the correct API (Converse or InvokeModel) without user configuration.

Built-in models:

• Updated models.json with 63 active models + 9 legacy models with (Legacy) label
• Runtime inference profile discovery auto-applies correct geo prefix per region
• StopSequences stripping for incompatible models (DeepSeek, GPT-OSS)
• Temperature auto-retry for models that reject the temperature parameter

Custom/imported models:

• Format auto-detection via InvokeModel probe at init time
• Supports imported, fine-tuned (custom-model-deployment), and provisioned-throughput ARNs
• ARN validation with actionable error messages
• Tested with 6 imported models across 4 architectures

Error handling:

• normalizeBedrockError() rewrites raw Bedrock errors into user-facing guidance
• Errors propagate to UI instead of being swallowed silently

[gemini-code-assist]

Code Review

This pull request significantly enhances the AWS Bedrock integration by consolidating support for built-in, imported, fine-tuned, and provisioned-throughput models into a single node. Key improvements include runtime inference profile discovery, automated request format detection for imported models, and robust error normalization. Feedback identifies several areas for refinement: correcting typos in the model catalog, adjusting validation logic to allow geo-prefixed inference profile IDs, ensuring that input values of zero for temperature and tokens are not incorrectly defaulted, and adding safety checks for nullish response bodies from the AWS SDK.

[gemini-code-assist]

This check prevents the use of geo-prefixed inference profile IDs (e.g., us.anthropic...), which are explicitly handled in the resolveBedrockModel utility (utils.ts:310). This makes that logic unreachable for the customModel input. If geo-prefixed IDs are intended to be supported, the validation should be updated to allow them.

[Ankit5467]

The ARN-only validation on customModel is intentional — the field is labeled "Custom Model ARN" and only accepts arn:aws:bedrock:... values (imported models, provisioned throughput, custom model deployments, inference profile ARNs).

Geo-prefixed inference profile IDs (e.g., us.anthropic.claude-sonnet-4-6) are not user-facing inputs — they're auto-applied internally by resolveBedrockModel() based on the model's inference_profile_geos in models.json and the selected region. The GEO_PREFIX_RE branch at utils.ts:310 is reachable through that auto-application path, not through customModel input.

For built-in models, users use the "Model Name" dropdown + "Region" dropdown and the correct profile is applied automatically.

[Ankit5467]

@HenryHengZJ

[HenryHengZJ]

thanks @Ankit5467 ! with this changes, will this still works?

[Ankit5467]

thanks @Ankit5467 ! with this changes, will this still works?

Yes @HenryHengZJ . # 6309 is built on top of # 5731's merged state. We import getAWSCredentialConfig() from awsToolsUtils.ts (added by # 5731) and pass the resolved credentials through all new code paths: imported model info lookup, inference profile discovery, and the BedrockImportedChat constructor. AssumeRole works end-to-end for both built-in and custom model invocations.