fix(tool): stable cache key for @tool(cache_results) independent of tool_hooks (#4387)

[raviraj-441]

Description

Fixes issue #4387 where @tool(cache_results=True) does not work as expected when tool_hooks is not set.

Root Cause

The cache key computation was inconsistent between cache lookup and cache storage:

• Cache lookup used empty entrypoint_args that start empty
• Cache storage used entrypoint_args after mutation with arguments.update()
• This caused different cache keys for read vs write operations, resulting in cache misses

Solution

• Added _get_cache_key_from_resolved_args() method for deterministic cache key generation
• Added _build_resolved_args() helper to build canonical arguments before mutations
• Modified execute() and aexecute() to compute cache key once from resolved args and reuse for both lookup and storage
• Cache key computation now filters framework arguments and sorts for consistency

Changes Made

1. Function.py changes:

   • Added _get_cache_key_from_resolved_args() method
   • Added FunctionCall._build_resolved_args() helper method
   • Modified cache key computation in execute() and aexecute() methods

2. Test coverage:

   • Added test_tool_cache_without_hooks() - verifies caching with no tool_hooks
   • Added test_tool_cache_with_empty_hooks() - verifies caching with empty tool_hooks
   • Added test_tool_cache_kwargs_order_independence() - verifies order independence

Testing

• ✅ All new tests pass
• ✅ Existing test suite passes (37/38 tests pass, 1 unrelated Windows path test fails)
• ✅ Manual testing confirms cache hits on second identical calls

Impact

• Fixes caching behavior for @tool(cache_results=True) when tool_hooks is not provided
• Maintains backward compatibility when tool_hooks is present
• Improves cache key consistency regardless of argument order