[Suggestion] Add context health analysis #36

Is your feature request related to a problem? Please describe.
Users can see token composition, but not how close the active model is to its context limit or when context growth is becoming dangerous.

Describe the solution you'd like
Show context used versus model limit, remaining headroom, growth over time, cache status, and estimated compaction risk in the sidebar and /analyze.

Additional context
Execution order: 4, after #35. Use model limits from provider metadata and label estimated versus exact values clearly.