test(usage-analysis): add unit tests for mergeUsageAnalysis, analyzeContextReferences, deriveConversationPatterns

[rajbos]

Summary

Adds 23 unit tests for the synchronous, pure-function core of usageAnalysis.ts -- the 1,974-line module driving the Usage Analysis dashboard that had zero test coverage.

What's tested

mergeUsageAnalysis (7 tests)

• Tool call counts accumulate correctly across multiple sessions (including per-tool byTool map)
• Mode usage (ask/edit/agent/plan/customAgent) accumulates correctly
• Context reference fields (file, workspace, codebase) accumulate correctly
• MCP tool counts accumulate by server and by tool
• Mixed-tier session tracking: sessions with modelCount > 0 update mixedTierSessions, standardModels, premiumModels, request counts
• Sessions with modelCount = 0 are correctly excluded from switching stats
• byKind context reference counts accumulate correctly

analyzeContextReferences (13 tests)

• Each of the 13 reference types detected: #file, #selection, @workspace, @Terminal, #codebase, #changes, #clipboard, #terminalLastCommand, #outputPanel, #problemsPanel
• Accumulates on top of existing counts
• Empty text produces no counts
• Matching is case-insensitive

deriveConversationPatterns (3 tests)

• 0 requests: singleTurn=0, multiTurn=0, avgTurns=0
• 1 request: produces single-turn session with avgTurns=1
• 3 requests: produces multi-turn session with correct avgTurns and maxTurns

Approach

Scoped to the purely synchronous functions only. analyzeSessionUsage and calculateModelSwitching are async with wide dependency interfaces and are left for a follow-up. Tests use inline object construction -- no mocks, no file I/O.