fix: OpenAI StructuredModeResolver should match versioned model names

[ipalaus]

Summary

StructuredModeResolver::supportsStructuredMode() uses in_array with exact string matching. Versioned model names like gpt-5-mini-2025-08-07 don't match gpt-5-mini, causing them to silently fall into JSON mode (json_object) instead of Structured mode (json_schema + strict: true).

In JSON mode the schema is embedded as plain text in a system message and json_object format is used — no constrained decoding. This results in significantly more verbose model output compared to proper structured mode.

Approach

Uses a hybrid strategy:

• Exact matching for gpt-4o and o3-mini families, where only specific versions support structured output (e.g., gpt-4o-2024-08-06 does, earlier versions don't)
• Prefix matching for gpt-4.1+, gpt-4.5+, and gpt-5+ families, where structured output is a baseline feature across all versions

This avoids the need to update the allowlist every time OpenAI releases a new dated version of a gpt-5 family model, while remaining conservative for older model families with inconsistent support.

Changes

• Split supportsStructuredMode() into exact matches (gpt-4o, o3) and prefix matches (gpt-4.1+, gpt-5+)
• Add StructuredModeResolverTest with 32 test cases covering exact matches, prefix matches, JSON mode fallback, and unsupported model exceptions

Test plan

• All existing tests pass (1495 passed, 8 skipped)
• 32 new tests for model name resolution
• gpt-5-mini-2025-08-07 → Structured mode (was JSON mode)
• gpt-5-mini → Structured mode (unchanged)
• gpt-4o-mini-2024-07-18 → Structured mode (unchanged, exact match preserved)
• gpt-4-turbo → JSON mode (unchanged)
• o1-mini → throws (unchanged)