fix(provider-health): disable thinking and tools for probes

[szafranski]

Summary

• disable model thinking for provider health probes
• use an empty tool policy so probes do not advertise agent tools
• keep normal conversation configuration unchanged

Why

A short health-check prompt could time out on thinking-enabled models because the probe still carried reasoning overhead and full agent tool definitions. A working provider was therefore reported as unhealthy.

Validation

• cargo test -p aionui-ai-agent - 870 passed, 12 ignored
• cargo clippy -p aionui-ai-agent -- -D warnings
• cargo fmt --all -- --check
• git diff --check
• just build-debug
• just push full gate - 7,996 passed, 22 skipped
• live health probe completed in 18.77 seconds within the existing 30-second timeout, with no advertised tools

Existing structured health-check logging is sufficient; this change does not add payload logging.

Fixes #744