fix(bedrock): disable thinking mode when forcing tool_choice

[strands-agent]

Description

Fixes #1485 - Cannot use structured_output with thinking enabled.

Supersedes #1490 (cleaner implementation rebased on latest main after #1336 pyupgrade merge)

Problem

When using BedrockModel with thinking mode enabled via additional_request_fields, if structured_output fails to generate a tool call on the first attempt, the event loop forces tool_choice (set to {"any": {}}) to retry. However, Bedrock's API doesn't allow thinking mode to be enabled when tool_choice forces tool use, resulting in:

ValidationException: The model returned the following errors: Thinking may not be enabled when tool_choice forces tool use.

Solution

Adds a helper method _get_additional_request_fields() that temporarily removes the thinking configuration from additionalModelRequestFields when tool_choice is forcing tool use (i.e., not {"auto": {}}).

Key changes:

1. Detect when tool_choice is forcing a tool (not auto and not None)
2. Create a copy of additional_request_fields without the thinking key when forcing
3. Preserve thinking mode for all normal agent interactions (when tool_choice is auto or None)
4. Preserve all other additional_request_fields when removing thinking

Testing

Added 7 comprehensive unit tests in test_bedrock_thinking.py covering:

• ✅ Thinking removed when forcing tool use ({"any": {}})
• ✅ Thinking removed when forcing specific tool ({"tool": {...}})
• ✅ Thinking preserved with auto tool choice ({"auto": {}})
• ✅ Thinking preserved when no tool choice (None)
• ✅ Other additional fields preserved when thinking is removed
• ✅ Models without thinking config work normally
• ✅ Fields without thinking preserved when forcing tool

Behavior

Before: Agent crashes with ValidationException when structured_output + thinking enabled
After: Agent gracefully retries structured_output by temporarily disabling thinking for that specific retry

Related Issues

Fixes #1485
Supersedes #1490

Type of Change

• Bug fix

Testing

• hatch test tests/strands/models/test_bedrock_thinking.py - 7 passed
• hatch test tests/strands/models/test_bedrock.py - 94 passed
• hatch fmt --linter - All checks passed

Checklist

• I have read the CONTRIBUTING document
• I have added tests that prove my fix is effective
• My changes generate no new warnings
• This is a non-breaking change (thinking mode still works for normal interactions)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!