perf(startup): add release timing harness

[limityan]

Summary

• Add a sanitized startup/session performance trace snapshot for debug and explicit perf E2E runs.
• Add a release-fast performance E2E path, a long-session fixture generator, and a startup/session timing spec.
• Enable embedded WebDriver for release-fast E2E through the existing devtools feature and update WebDriver routes to Axum 0.8 path syntax.
• Stabilize the startup splash background with a shared --bitfun-startup-bg variable so light-theme startup no longer flashes from a dark splash fallback to a white splash.

Refs #949

Performance Data

Scenario: Windows local release-fast build with production web bundle, release-fast Rust profile, embedded WebDriver, 80 generated session metadata rows, and one 80-turn long session with 160 synthetic tool results.

Latest post-rebase sample from this PR:

Scenario	Metric	Result
Startup	first_script_eval	1358.4 ms
Startup	before_render duration	32.7 ms
Startup	app_effect_mounted	1484.0 ms
Startup	main_window_shown	1484.6 ms
Startup	non_critical_init_done	1886.8 ms
Startup	interactive_shell_ready	1943.5 ms
Long session first open	initially loaded turns	3 / 80
Long session first open	restore latest turns	39.4 ms
Long session first open	state commit	2.5 ms
Long session first open	latest-turn first frame	48.9 ms
Long session background hydrate	full hydrate	56.5 ms

Startup API aggregate in the same sample:

Command	Count	Total	Max
get_config	10	1438.7 ms	357.5 ms
initialize_ai	1	335.9 ms	335.9 ms
git_is_repository	4	73.3 ms	28.7 ms
remote_get_workspace_info	1	69.5 ms	69.5 ms
git_get_repository_basic	3	51.8 ms	17.8 ms

Earlier repeat samples on the same release-fast branch before the final upstream rebase showed startup variance rather than a stable startup reduction:

Scenario	Metric	Min	P50	Max
Startup, 3 runs	interactive_shell_ready	1438.1 ms	1554.0 ms	1597.2 ms
Startup, 3 runs	main_window_shown	1066.2 ms	1098.9 ms	1188.7 ms
Long session, 3 runs	latest-turn first frame	36.0 ms	49.5 ms	64.5 ms

One earlier cold-ish long-session background hydrate outlier hit about 1535.9 ms; backend logs attributed it to reading 80 turn files with about 1.9 MB raw tool-result text. The perceived first frame still stayed near the 50 ms range because only the latest 3 turns are restored initially.

This PR should be evaluated as measurement infrastructure plus startup visual consistency. It does not claim an end-to-end startup reduction; the data points to follow-up optimization work around startup API duplication and early initialization.

Risks / Notes

• The diagnostic window.__BITFUN_STARTUP_TRACE__ surface exposes only sanitized phase/API aggregates, not raw request payloads or workspace paths, and is limited to dev or explicit perf WebDriver runs.
• devtools now enables embedded WebDriver for release-fast E2E. End-user release builds are unaffected unless that feature is explicitly enabled.
• Session nav data-testid attributes are test-only selectors; data-session-title mirrors already visible UI text.
• The long-session fixture writes synthetic local test data and requires an explicit workspace path when run.
• Startup splash background is now tied to the earliest known theme background; if a platform shows a native pre-document blank window before web content, that remains a separate native show-timing problem.

Verification

• pnpm run type-check:web
• pnpm --dir src/web-ui run test:run src/shared/utils/startupTrace.test.ts src/infrastructure/theme/core/ThemeService.test.ts
• pnpm run check:repo-hygiene
• cargo check -p bitfun-desktop --features devtools (passes with existing parse_clipboard_path_segments unused warning)
• pnpm run desktop:build:release-fast (passes with existing Vite chunk warnings and the same Rust warning)
• E2E_TEST_WORKSPACE=<workspace> BITFUN_E2E_PERF_SESSION_ID=perf-long-session-000 pnpm run e2e:test:perf:release-fast
• git diff --check gcwing/main...HEAD