fix(models): forward generation config parameters in AnthropicLlm

[sebastienc]

Summary

Fixes #5393

Disclaimer: I'm using Claude Code to generate the following text. I find the summary is better than what I would've wrote.

AnthropicLlm was silently ignoring temperature, top_p, top_k, and stop_sequences from GenerateContentConfig. This change wires them through to both the streaming and non-streaming messages.create calls, and adds output_config.effort support via a new AnthropicGenerateContentConfig escape hatch and a ThinkingLevel → effort mapping.

• temperature / top_p / top_k / stop_sequences — forwarded to messages.create in both streaming and non-streaming paths; each is guarded by is not None so unset params remain NOT_GIVEN rather than None
• Sampling params disabled when effort is set — temperature, top_p, and top_k are suppressed (sent as NOT_GIVEN) when output_config.effort is active, since Anthropic does not allow sampling parameters alongside effort-based generation
• _build_effort_param — maps ThinkingConfig.thinking_level to Anthropic's output_config.effort string (MINIMAL → "low", LOW → "medium", MEDIUM → "high", HIGH → "xhigh"); also reads AnthropicGenerateContentConfig.effort directly for cases needing "max"
• AnthropicGenerateContentConfig — subclass of GenerateContentConfig that adds a single effort field accepting "low" | "medium" | "high" | "xhigh" | "max"; callers who need the "max" effort level (not expressible via ThinkingLevel) can pass this config instead

Testing plan

12 new tests added to tests/unittests/models/test_anthropic_llm.py. Full suite: 71 passed.

tests/unittests/models/test_anthropic_llm.py::test_non_streaming_forwards_generation_params PASSED
tests/unittests/models/test_anthropic_llm.py::test_non_streaming_omits_unset_generation_params PASSED
tests/unittests/models/test_anthropic_llm.py::test_streaming_forwards_generation_params PASSED
tests/unittests/models/test_anthropic_llm.py::test_streaming_omits_unset_generation_params PASSED
tests/unittests/models/test_anthropic_llm.py::test_thinking_level_maps_to_effort[MINIMAL-low] PASSED
tests/unittests/models/test_anthropic_llm.py::test_thinking_level_maps_to_effort[LOW-medium] PASSED
tests/unittests/models/test_anthropic_llm.py::test_thinking_level_maps_to_effort[MEDIUM-high] PASSED
tests/unittests/models/test_anthropic_llm.py::test_thinking_level_maps_to_effort[HIGH-xhigh] PASSED
tests/unittests/models/test_anthropic_llm.py::test_anthropic_config_effort_max_non_streaming PASSED
tests/unittests/models/test_anthropic_llm.py::test_anthropic_config_effort_max_streaming PASSED
tests/unittests/models/test_anthropic_llm.py::test_no_effort_when_thinking_level_unset PASSED
tests/unittests/models/test_anthropic_llm.py::test_effort_takes_priority_over_sampling_params PASSED

[sebastienc]

Fixed the pyink issue and rebased.