Improve loading speed of dashboard results

[micahjo7]

Dashboard has been slow to load GCS results after login, and slow loading the individual suite and test case/task results. Improves speed with caching and pagination with 20 results per page.

Improves initial load speed from around 6s to 1s, and speed of suite/test results displaying from ~3s to <1s.

Summary of Performance Improvements

1. Sub-Second Click Latency & Accordion Optimizations (dashboard.js / api.js)

• Suite Directory Pre-fetching (getSuiteDirectoryListing): Introduced a recursive suite directory prefix pre-fetcher triggered immediately when a dashboard loads. Instead of firing up to 6 serial/parallel GCS network listings upon expanding a task accordion, all runs' index points, logs, and trajectories are now resolved completely synchronously in local memory from the pre-fetched list.
• Persistent Prompt & BaseApp Caching: Implemented IndexedDB caching for the resolved promptText and baseApp of each scenario under the key prompt:${testId}:${scenarioName}. Expanding an accordion a second time skips all file reading and parsing entirely, leading to 0ms click latency.
• Persistent Text Download Caching (getFileText): Added caching for raw text downloads (e.g. log files, run.mjs scripts) to avoid repeat GCS queries, returning cached contents instantly.
• Local File 404 Check Bypass: Pre-fetched recursive file listings are queried locally in memory to verify file existence instead of using prefix-checking network roundtrips (_checkRemoteFileExists), saving ~200-300ms per file probe.

2. High-Speed Numbered Pagination (landing.js / landing.css / index.html)

• Reverse-Chronological Pagination: Unified suite listings on the landing page to display exactly 20 items per page. Evals metadata (evals.json) is loaded only for the visible 20 suites, accelerating initial remote loads from ~10 seconds to <300ms.
• IndexedDB Metadata Preloading (preloadCachedSuitesMetadata): Scans all previously cached remote suite metadata in parallel on startup (IndexedDB local query takes <5ms total), enabling complete dataset filtering and sorting.
• Full-Dataset Filtering & median statistics: Redesigned filtering to calculate filters across the entire dataset before slicing for pagination (rather than filtering only the current page's 20 visible items). The median uplift and dumbbell charts at the top now represent the entire filtered dataset.
• Premium Page-Number Jump Controls: Replaced the simple text pagination label with a premium page-number navigation strip.
  • Automatically handles ellipses (...) to span large page boundaries (e.g., 1 ... 4 [5] 6 ... 12).
  • Boundary navigation buttons (Previous / Next) are HTML disabled and styled with .secondary-btn:disabled (opacity 35%, cursor not-allowed, pointer-events disabled) on the first and last pages.

[paulirish]

I'm holding off on this cuz it feels like a lot of complication and there are probably some more straightforward fixes.
But i dont have time right now to look at the details.

[micahjo7]

closed in favor of #1179