perf(startup): bootstrap workspace startup state

[limityan]

Summary

• Prepare the workspace startup state snapshot during desktop startup and inject it into the WebView bootstrap script so the first window can avoid the startup initializeWorkspaceStartupState IPC waterfall.
• Keep the original frontend command as the fallback path when the bootstrap snapshot is missing, invalid, cannot be serialized, or exceeds the 64KB bootstrap cap.
• Add bounded E2E-only resource timing data to startup performance reports for future cold-start bottleneck analysis without adding runtime log noise.

Refs #949

Performance data

Test profile: release-fast, 5 cold-start samples. Baseline is latest gcwing/main at d8f2d2e2; modified result is this PR at 4e1f4e27.

Scenario / metric	Before	After	Delta
Cold start: shell interactive median	450.1 ms	442.6 ms	-7.5 ms
Cold start: first script eval median	223.9 ms	223.5 ms	-0.4 ms
Cold start: window shown to interactive median	187.5 ms	186.2 ms	-1.3 ms
Workspace startup state available to frontend	118.6 ms	7.4 ms	-111.2 ms
Workspace startup state step	114.8 ms	4.6 ms	-110.2 ms
initializeWorkspaceStartupState frontend IPC wait	113.7 ms	not on first-window path	removed from first-window path
Native bootstrap snapshot preparation	n/a	1.0 ms	+1.0 ms before WebView creation
Native create_main_window median	497 ms	491 ms	-6 ms
Native WebView build median	458 ms	450 ms	-8 ms

Long-session interaction guard after the change, using the focused release-fast core profile:

Guard scenario	Result
First open latest text visible	183.3 ms
First open full history ready	251.3 ms
Rapid switch latest text visible	151.0 ms
Rapid switch latest usable	187.8 ms
First scroll latest text visible	174.4 ms
Resize latest text visible	154.9 ms
Blank surface events across guard samples	0

Risk and compatibility

• This moves existing workspace startup state preparation before the main WebView is created. The measured native preparation median is 1.0 ms; the 64KB cap prevents large workspace lists from being injected into the bootstrap script. Oversized or failed snapshots fall back to the existing IPC command.
• The implementation reuses the existing startup-state collection path instead of introducing a second data model, so workspace cleanup/current workspace/recent workspace/legacy remote snapshot semantics stay aligned with the command path.
• Remote workspace data is not transformed differently by this PR. The frontend maps the injected snapshot through the same mapWorkspaceStartupStateSnapshot logic used by the command response.
• Resource timing is collected only inside the E2E startup report and is bounded to 40 sampled entries per list. It does not add product runtime logging.

Validation

• pnpm --dir src/web-ui run test:run src/shared/types/global-state.test.ts src/app/startup/startupPerformanceContract.test.ts
• pnpm run type-check:web
• cargo check -p bitfun-desktop
• node tests/e2e/scripts/perf-coverage-contract.test.mjs
• git diff --check gcwing/main..HEAD
• pnpm run e2e:test:perf:startup-stability:release-fast
• pnpm run e2e:test:perf:long-session-interactions:release-fast -- --profile core