Adds tracking of actual token usage per model.

[davidbreitgand]

What does this PR do?

1. Adds tracking of actual token usage per model
2. Hardens implementation: protects against an accidental writing to the nil datastore
3. Current implementation is limited to OpenAI as Phase 1, other providers will be added in subsequent PRs

Why is this change needed?

requestmetadata/plugin.go extracts max_tokens on the inbound request event. max_tokens is optional and is only a proxy for the actual token usage. The actual token usage will be instrumental for advanced cost-aware scoring, accurate cost tracking to capture price reversals, cache management, and visualization.

How was this tested?

• Unit tests added/updated
• Integration/e2e tests added/updated
• Manual testing performed

Checklist

• Commits are signed off (git commit -s) per DCO
• Code follows project contributing guidelines
• Tests pass locally (make test)
• Linters pass (make lint)
• Documentation updated (if applicable)

Related Issues

Fixes #115

[davidbreitgand]

@shmuelk , @Mohammad-nassar10 please review

[davidbreitgand]

/cc @shmuelk
/cc @Mohammad-nassar10

[davidbreitgand]

/cc @ronenkat

@ronenkat : thanks for the feedback on the counters overflow. Fixed. Please take another look.