fix: update model aliases and enhance event parsing for Copilot provider

[NihalJain]

• Fixes Copilot: Models like claude-sonnet-4.6 are classified as Copilot (auto) #219

[iamtoruk]

Thanks for the PR @NihalJain, good find on the model classification issue. Reviewed the changes in detail:

Model aliases in models.ts -- These are correct and needed. claude-sonnet-4.6 (dot-version, tier-first) is a different format from what getCanonicalName produces, so explicit aliases are the right approach.

Catch-all type on LegacyCopilotEvent -- Good, consistent with the transcript format which already has this.

toolu_ prefix in tool call hints -- Correct and properly ordered after the more specific toolu_bdrk_ and toolu_vrtx_ prefixes, so specificity is preserved.

Two things I'd like adjusted before merging:

1. The data.model extraction in parseLegacyEvents (lines 89-92 of the diff) -- Casting event.data broadly and extracting model from ANY event type is fragile. If a future event type has a model field meaning something different (e.g. a "3D model" reference), it would silently corrupt currentModel. Safer to scope this to specific event types, or at minimum only apply it when currentModel is still empty (i.e. as a fallback, not an override).

2. Fallback change from 'copilot-auto' to COPILOT_OPENAI_AUTO -- This changes cost attribution from Claude pricing ($3/$15 per MTok) to OpenAI GPT pricing when the model can't be inferred. This is a meaningful behavioral change. copilot-auto was intentionally a neutral catch-all. If a user is on Anthropic models with no tool calls (so no ID prefixes to infer from), their costs would now be calculated at OpenAI rates. Can you share the reasoning here? If the intent is "Copilot defaults to OpenAI", that's reasonable for the fallback in inferModelFromToolCallIds, but should be paired with a similar default in parseLegacyEvents for consistency.

The data.model extraction in inferModelFromToolCallIds with weight 100 is fine -- explicit model IDs should outweigh heuristic inference.

[iamtoruk]

Retracting my earlier comment -- I jumped to review without checking with you first. Please disregard the previous feedback; we'll follow up separately once we've aligned internally on what we want from this PR.