feat: set up gen ai inference attributes for foundational text models in java v1 sdk

[yiyuan-he]

Description of changes:
Adding auto-instrumentation support for GenAI inference parameters for Java V1 AWS SDK.

The following foundational text models are supported:

• AI21 Jamba
• Amazon Titan
• Anthropic Claude
• Cohere Command R
• Meta Llama
• Mistral AI

Full list can be found here.

Note: Removed support for old Cohere Command models since they already throw a 404 response for Java V1.

Note: There is a backwards compatibility issue with the existing tests as they rely on older versions of the AWS SDK. Since Bedrock is newer resource we are forced to introduce a newer SDK version which implicitly adds in behavior that is not backwards compatible with the existing tests. It seems these dependency issues in the tests are a known issue in upstream which is why they rely on such old AWS resource versions in the first place. Synced with @mxiamxia and we decided to temporarily disable these tests for now as we figure out a long-term solution. We rely on the contract tests to verify the SDK quality for the time being.

New inference parameter attributes added according to OpenTelemetry Semantic Conventions for GenAI attributes:

• gen_ai.request.max_tokens
• gen_ai.request.temperature
• gen_ai.request.top_p
• gen_ai.response.finish_reasons
• gen_ai.usage.input_tokens
• gen_ai.usage.output_tokens

Test Plan:
Set up sample app to make Bedrock Runtime InvokeModel API calls to the supported foundational models and verified the auto-instrumentation attributes.

AI21 Jamba

Amazon Titan

Anthropic Claude

Cohere Command

Meta Llama

Mistral AI

[yiyuan-he]

Deprecating in favor of new PR