Improve startup and historical session readiness

[limityan]

Summary

This PR improves startup observability and historical session readiness without changing chunkSizeWarningLimit and without submitting design or baseline documents.

Changes

• Add lightweight startup trace aggregation for desktop/web startup phases and API payload/remote-call counters.
• Reduce pre-render startup blocking by moving non-critical context/recommendation/config watcher setup after the first render path.
• Dedupe concurrent ConfigManager.getConfig() reads for the same path to avoid repeated IPC during startup bursts.
• Track historical session hydrate state explicitly (metadata-only, hydrating, ready, failed) so opening a saved session shows a loading/retry state instead of briefly looking like a new empty session.
• Reuse pending historical hydrate before ensuring backend session setup, reducing duplicate session restore/create work.
• Add focused regression tests for startup tracing, config read dedupe, history hydrate transitions, failed/retry UI, pending hydrate reuse, ACP restore skip, and Modern store sync preservation.

Performance Results

Local release-like startup sampling, 3 runs each:

Metric	Before	After	Result
First Render Scheduled avg	82.8ms	68.0ms	-17.9%
Start Application avg	1249.3ms	1038.5ms	-16.9%
After Render avg	1165.1ms	969.4ms	-16.8%
Main Window Shown avg	2285.3ms	2004.4ms	-12.3%
Interactive Shell Ready avg	2447.2ms	2196.3ms	-10.3%
Startup API count avg	21.0	20.0	-1 call
Startup response bytes avg	14561	13559	-1002 bytes
Remote command count avg	0.0	0.0	no local increase

Historical-session UX improvement: saved sessions now enter a visible loading/retry lifecycle immediately, instead of falling through to the new-session empty state while turns are hydrating.

Risks / Follow-ups

• Native window creation had one outlier in the after sample; this PR does not modify native window creation, so that metric is not claimed as a direct win and should continue to be sampled.
• Real remote SSH workspace sampling was not available locally. The trace counters show no local remote-call increase, but remote command count / payload / serialization data should be verified with a real remote workspace before making remote-specific claims.
• Deferring non-critical setup after render is intentionally conservative, but very early consumers of context/recommendation/config watcher state should be watched in manual smoke testing.
• Historical sessions still load full turns; this PR avoids high-risk paging/worker/protocol changes. Very large sessions may still need a later, separately reviewed optimization.
• Existing Vite chunk-size / dynamic-import warnings remain. This PR does not hide them by raising thresholds.

Validation

• git diff --cached --check
• rustfmt --edition 2021 --check --config skip_children=true src\apps\desktop\src\lib.rs src\apps\desktop\src\theme.rs
• pnpm run lint:web
• pnpm run type-check:web
• pnpm --dir src/web-ui run test:run (124 files / 650 tests)
• cargo check -p bitfun-desktop
• cargo test -p bitfun-desktop (27 tests)
• pnpm run desktop:build:release-fast passed during local milestone verification; existing Vite chunk warnings are still present.