fix: disable automatic cache affinity for Azure OpenAI

[koriyoshi2041]

Problem

With the openai provider and an *.openai.azure.com base URL, automatic cache affinity currently adds OpenAI's prompt_cache_key field. Azure Foundry rejects that field with unrecognized_request_argument, breaking LLM-backed operations.

Fixes #3518.

Fix

Remove Azure OpenAI hosts from the automatic prompt_cache_key allowlist. Native OpenAI and openai.com hosts keep the existing behavior, while Azure hosts now resolve auto to none.

The configuration and Azure setup docs now describe the same behavior.

Test

• uv run pytest tests/test_cache_affinity.py -q — 67 passed
• Ruff check and format check on the changed Python files — passed
• uv run ty check hindsight_api/engine/cache_affinity.py hindsight_api/config.py — passed
• git diff --check — passed

The repository-wide lint script reached the TypeScript lint step but could not load @eslint/js in this fresh worktree. No TypeScript files are changed by this PR.

Risk

Low. This only narrows the auto host allowlist. Operators who explicitly select openai_prompt_cache_key retain that override, and non-Azure host behavior is unchanged.