feat: introduce fine_tuning API

[eoinfennessy]

What does this PR do?

Closes #3708

Scaffolds the fine_tuning API. This is intended to be an OpenAI compatible API that may allow us to deprecate our post-training API. No providers have been implemented / migrated. Current conformance summary is provided below.

OpenAI Fine-Tuning API Conformance Summary

Conformance Score: 88.1% (63 schema issues, 1 missing property, 0 breaking changes)

✅ Implementation Status

6 Endpoints Implemented:

• POST /v1/fine_tuning/jobs - Create fine-tuning job
• GET /v1/fine_tuning/jobs - List fine-tuning jobs
• GET /v1/fine_tuning/jobs/{id} - Retrieve fine-tuning job
• POST /v1/fine_tuning/jobs/{id}/cancel - Cancel fine-tuning job
• GET /v1/fine_tuning/jobs/{id}/checkpoints - List job checkpoints
• GET /v1/fine_tuning/jobs/{id}/events - List job events

OpenAI-Compatible Models:

• FineTuningJob - Core job object with all OpenAI fields
• FineTuneMethod - Supports supervised, DPO, and reinforcement methods
• FineTuningJobHyperparameters - Training configuration with auto defaults
• FineTuningJobEvent - Job logging and events
• FineTuningJobCheckpoint - Training checkpoints

🔧 Key Technical Details

Nullable Field Handling:

• Applied nullable_openai_style for required nullable fields (e.g., finished_at, error, validation_file)
• Applied remove_null_from_anyof for optional non-nullable fields (e.g., hyperparameters, seed, organization_id)
• Ensures OpenAPI 3.0 compatibility while maintaining functional equivalence

Breaking Changes: None (verified with oasdiff)

📊 Remaining Issues (Non-Breaking)

The 63 remaining issues are format/representation differences that don't affect API functionality:

• ~30 issues: Nullable field format differences (anyOf vs nullable: true - functionally equivalent)
• ~12 issues: Hyperparameter union representation differences
• ~10 issues: Default value improvements (better UX)
• ~8 issues: Optional field representation
• ~3 issues: Other minor format differences

1 Missing Property: metadata query parameter on GET /fine_tuning/jobs (FastAPI doesn't support deepObject style parameters natively)

🎯 Conformance Strategy

Prioritized functional compatibility over strict format matching:

• All request/response types match OpenAI's structure
• No breaking changes for API consumers
• Format differences are benign (different valid representations of the same schema)

[mergify]

This pull request has merge conflicts that must be resolved before it can be merged. @eoinfennessy please rebase it. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork