PR6: measured repository and UI cleanup

[VolkerChristian]

Summary

• replace repeated filesystem-watch membership scans with set-based reconciliation
• preserve repository selection, Inspector snapshots, and settings catalog behavior where profiling did not justify change
• humanize machine-oriented event labels only at UI render boundaries
• consolidate related component-local styling into the canonical application stylesheet
• add focused rename and display-boundary regression coverage

Measurements

Release microbenchmarks, median timings:

Area	Input	Before	After / candidate	Decision
Watch reconciliation	100 paths	24.9 us	3.0 us	Implemented; 8.2x faster
Watch reconciliation	1,000 paths	1.90 ms	0.043 ms	Implemented; 44x faster
Repository resolution	8 repos, 8 hints	13.24 ms	12.79 ms	Retained heuristic; 3.4% was not worthwhile
Repository resolution	32 repos, 32 hints	175.0 ms	164.1 ms	Retained heuristic; diff work dominates
Inspector unchanged refresh	10 / 100 / 1,000 agents	0.004 / 0.040 / 0.423 ms	No rebuilding observed	No optimization
Catalog insertion	10 models	9.05 us	7.59 us	Representative gain was negligible

Simplification and behavior

• removed quadratic QStringList::contains reconciliation and the separate duplicate-removal pass
• removed the local settings friendly() formatter in favor of one shared display helper
• removed three component-local style paths in favor of canonical application styling
• kept raw protocol values unchanged; for example, contextCompaction renders as Context compaction
• retained deleted-file, parent-directory, rename, recreation, atomic-write, and untracked-file behavior
• introduced no controller, repository, cache, event system, or duplicated state

Tests

• focused conversation cards, application layout, and live Git changes: 3/3 passed
• focused socket-pair and shell integration: 2/2 passed
• complete suite: 7/7 passed

The SNode transport tests require execution outside the managed syscall sandbox; inside it, the event loop receives EPERM. No transport code change was necessary.

Reviewer verdicts

• performance: pass
• filesystem behavior: pass
• protocol-value preservation: pass
• scope and exclusions: pass
• test completeness: pass

Conversation-card composition/refactoring remains reserved for PR7.