fix(models/anthropic): eliminate ParsedTextBlock Pydantic serialization warnings

[FadelT]

Problem

When using strands-agents >= v1.29.0 with anthropic SDK >= v0.84.0, every agent invocation using AnthropicModel produces 6+ Pydantic serialization warnings on stderr. The Anthropic SDK returns ParsedTextBlock objects in streaming responses, but calling event.model_dump() triggers serialization warnings because ParsedTextBlock doesn't match the discriminated union schema.

Impact:

• Affects all Strands users with anthropic >= 0.84.0
• Produces noisy stderr output (6+ warnings per agent call)
• Particularly problematic for demos and getting-started experiences
• Creates confusion - developers may think their code is broken

Closes #1865

Solution

Extract a _build_event_dict() helper method that constructs event dicts by directly accessing typed attributes instead of calling model_dump(). This approach:

1. Follows proven pattern - Same technique as the existing message_stop fix from [BUG] PydanticSerializationUnexpectedValue warnings when Anthropic SDK returns ParsedTextBlock in message content #1746
2. Reduces complexity - Extracts nested conditional logic into a focused helper
3. Maintains compatibility - No changes to public API or behavior
4. Complete coverage - Handles all event types: message_start, content_block_start, content_block_delta, content_block_stop, message_stop

Why this approach:
The Anthropic SDK's ParsedTextBlock objects have all required attributes accessible directly. Building dicts from these typed attributes bypasses Pydantic's serialization layer entirely, eliminating the warnings without changing SDK contracts.

Verification

Real SDK Testing (anthropic==0.109.2)

Tested with actual Anthropic API calls using claude-sonnet-4-6:

Before fix: Multiple Pydantic warnings on stderr
After fix: Zero warnings ✓

Test coverage:

• ✅ Text streaming - no warnings
• ✅ Tool-use streaming - no warnings
• ✅ All event types exercised

Unit Tests

• ✅ 59/59 tests passing
• ✅ Added 2 comprehensive warning-detection tests:
  • test_stream_text_no_pydantic_warnings
  • test_stream_tool_use_no_pydantic_warnings
• ✅ Updated existing test mocks to match new implementation

Changes

src/strands/models/anthropic.py

• Add _build_event_dict() static method - constructs dicts without Pydantic serialization
• Refactor stream() to use helper for all event types (reduces cognitive complexity)

tests/strands/models/test_anthropic.py

• Fix mock objects to use proper attribute assignment (not constructor kwargs)
• Add comprehensive warning-free streaming tests for text and tool-use
• Update test_structured_output mocks to match new pattern

Checklist

• Code follows project style guidelines
• Tests added/updated and passing (59/59)
• Verified with real Anthropic SDK (0.109.2)
• No breaking changes
• Follows pattern from [BUG] PydanticSerializationUnexpectedValue warnings when Anthropic SDK returns ParsedTextBlock in message content #1746
• Documentation updated (inline docstrings)

References

• Fixes [BUG] Strands Agent AnthropicModel emits Pydantic serialization warnings with Anthropic SDK v0.84.0 #1865
• Related to [BUG] PydanticSerializationUnexpectedValue warnings when Anthropic SDK returns ParsedTextBlock in message content #1746 (message_stop fix)