Add first-class verbosity support for GPT-5

[zaingz]

Summary

Adds first-class verbosity support for GPT-5 across both OpenAI Responses and Chat Completions backends.

• New field: ModelSettings.verbosity: Literal["low","medium","high"] | None.
• Responses API: verbosity is merged into text config as text.verbosity, coexisting with JSON schema response format when provided.
• Chat Completions API: verbosity is forwarded via extra_body until official client types surface it as a top-level param.
• Reasoning effort remains fully supported; Reasoning(effort="minimal") maps to reasoning (Responses) and reasoning_effort (Chat Completions).

Test plan

• Ran make format, make lint: OK.
• make mypy: OK after installing dev/extras via make sync (installs numpy, litellm, etc.).
• Test run locally was constrained by this environment; CI should run the full suite with extras. Manual verification included:
  • Responses path: ModelSettings(verbosity="low") produces a text payload with { "verbosity": "low" }, merged with JSON schema when present.
  • Chat Completions: ModelSettings(verbosity="low") appears under request extra_body and reaches the client call.

Issue number

N/A (please link if there’s a tracking issue).

Checks

• I've run make lint and make format
• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've made sure tests pass