feat(agents): add ratio history trimmer + history_processors wiring

[sanzog03]

Summary

Adds history_processors=[...] on super().init() alongside the existing capabilities=[...] spread, backed by a new
_build_history_processors_from_scalars() builder and a make_ratio_trimmer factory.

What it does

• PydanticAIBaseAgent now respects config.enable_trimming + config.trim_ratio — set them, and the agent automatically installs a ratio-based history trimmer that drops the oldest fraction
  of messages on every model request while preserving the first message (usually the system prompt).
• Consumers can also supply their own history processors directly via config.history_processors (new pass-through list, same shape as config.capabilities).
• Trimming stays opt-in by default. PydanticAIBaseAgentConfig overrides BaseAgentConfig.enable_trimming to False — the naive ratio trimmer can strand ToolReturnParts paired with earlier
  ToolCallParts, which OpenAI's API rejects. Safe opt-in once the caller knows their history shape is OK.
• Subclass-friendly: _build_history_processors_from_scalars follows the same override + super() pattern as _build_capabilities_from_scalars.

Files Changed:

• akd_ext/agents/_base/pydantic_ai/_capabilities.py — make_ratio_trimmer factory + module docstring noting the mode of future additions.
• akd_ext/agents/_base/pydantic_ai/_base.py — imports make_ratio_trimmer; adds history_processors: list[Any] and enable_trimming: bool = False override on PydanticAIBaseAgentConfig; adds
  the history_processors=[...] kwarg in super().init(); adds _build_history_processors_from_scalars() next to _build_capabilities_from_scalars.
• tests/agents/test_base_pydantic.py — adds seven tests under a new "make_ratio_trimmer" section.

Testing

Hermetic test suite (no network, no model keys):

uv run pytest tests/agents/test_base_pydantic.py -n=3

[NISH1001]

I was thinking of leaning towards:
https://github.com/vstorm-co/summarization-pydantic-ai