[perf] Webapp load-time wins (gzip, route-split, echarts tree-shake, compare cache)

[VforVitorio]

Webapp efficiency audit (2026-07-19) — the highest-leverage, lowest-risk wins to make loads/interactions more fluid. Ordered by impact. Full analysis in the audit notes.

• nginx serves the JS bundle uncompressed — webapp/nginx.conf has no gzip on; (stock nginx:alpine ships it off), so prod users download 1.88 MB instead of ~612 KB. Add gzip on; + gzip_types. ~67% less transfer, 1-line, zero risk. Do first.
• Route-level code-splitting — src/app/router.tsx imports every page eagerly; Home downloads every tab. React.lazy() per route + a Suspense boundary → ~300-450 KB gzip off first paint.
• ECharts full import — echarts-for-react default does import * as echarts (~1 MB). Switch chart features to echarts/core + echarts-for-react/core + explicit use([...]) (line/gauge/scatter/custom + 7 components) → ~180-220 KB gzip. Pair with manualChunks so echarts is one shared async chunk.
• Backend: compare Q-session path bypasses the warm session cache — backend/api/v1/endpoints/comparison.py:91-92 calls raw fastf1.get_session(...).load() instead of session_cache.get_loaded_session(...). This is the real cause of the ~14s cold compare. 2-line fix.
• Prewarm the compare session on 2nd-driver pick (like prewarmTelemetry) so the fetch is warm by COMPARE.

Already good (do NOT change): the replay rAF clock, staleTime:Infinity, useLapTelemetries parallel-fetch + signature-memoization, the prewarm pattern, the backend session_cache LRU, nginx hashed-asset cache headers. @tanstack/react-table/react-virtual are already tree-shaken out of the shipped bundle (dead weight in package.json only).

[VforVitorio]

PR #141 lands 3 of the 5 items — the highest-impact, lowest-risk ones:

• nginx gzip (~67% less transfer)
• route-level code-splitting (Home first paint ~611KB gzip → ~100KB; ECharts now a lazy chunk)
• backend: cache the compare Q-session load (the real cause of the slow cold compare)

Still open (deferred deliberately):

• ECharts echarts/core + explicit use([...]) tree-shake — the route-split already made ECharts lazy, so this is now a ~180KB trim on an already-lazy chunk. It's a delicate change across ~7 chart components (a missing component silently breaks a chart at runtime), so it wants a focused, all-charts-verified pass on its own.
• Prewarm the compare session on the 2nd-driver pick.