fix: preserve explicit auto-continue provider setting

[Brian Krabach (bkrabach)]

What changed

• Preserve the settings-only auto_continue override alongside extra_request_params when existing provider configuration is rewritten.
• Keep the allow-list narrow: preserve exact values, including native false and legacy string "false"; do not inject a default when absent or retain unrelated unknown keys.
• Document the manual settings override and its runtime-default behavior.

Why

The OpenAI and vLLM setup prompt is being removed while their runtime default remains enabled. Without this CLI preservation, users who manually opted out could have that setting silently dropped during provider edit/manage/replacement. This is the first dependency in the publication sequence.

How to verify

• uv run pytest tests/test_provider_config_reserved_keys.py — 35 passed.
• Broader pre-publication CLI suite: 2,548 passed (one platform-specific skip; integration tests excluded by the normal unit-suite marker).
• Installed-package acceptance: 50 cases passed across both providers and the real CLI configuration flow.

Breaking changes

None. The key remains optional and absent settings continue to defer to provider runtime defaults.