Process thinking blocks for Anthropic with Bedrock provider

[ferenci84]

Description

This is a continuation of PR #4426 with implementation for Bedrock.

Replacement of PR #4444

Checklist

• [] The relevant docs, if any, have been updated or created
• [] The relevant tests, if any, have been updated or created

(No docs have been updated because I believe more has to be done before this feature becomes public)

Testing instructions

It can be tested the same way as the PR 4426 with the following config:

{
      "title": "AWS - Claude 3.7 Sonnet (reasoning)",
      "provider": "bedrock",
      "model": "arn:aws:bedrock:us-east-1:874087778966:inference-profile/us.anthropic.claude-3-7-sonnet-20250219-v1:0",
      "region": "us-east-1",
      "profile": "default",
      "completionOptions": {
        "temperature": 1,
        "topP": 0.999,
        "maxTokens": 8192,
        "reasoning": true,
        "reasoningBudgetTokens": 1024,
      }
    }

[netlify]

✅ Deploy Preview for continuedev canceled.

Name	Link
🔨 Latest commit	7949f9a
🔍 Latest deploy log	https://app.netlify.com/sites/continuedev/deploys/67d0a710d5d6b30008c3dc68

[ferenci84]

@chezsmithy I created an issue for tool support: #4470, since you were working on it with bedrock, would you mind to check it?

[ferenci84]

@sestinj Can you look into this? Until we don't have a shared implementation, this PR contains solution for extended thinking for both the Anthropic and Bedrock provider, and it's now working well with tools.

[chezsmithy]

@ferenci84 and @sestinj: I've reviewed this locally and tested it against Claude 3.7 with thinking enabled on AWS Bedrock and it is functioning as expected. Great work!