Results pipeline rework (integration branch) - #2402

Ask for task totals instead of the full task list (needs yaptide/yaptide#1681), stop polling hidden tabs, find the current simulation by its input, and let failed cache fetches reject instead of hanging.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Pages kept in object storage come from the backend as signed links; they are
fetched in parallel, without credentials, and converted like the API
responses. Pages still sent inline are used as before.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…r-sent events (#2440)

- the simulation card shows what the partial results cover ("37% of primaries, 2 minutes ago") and a
  Partial Results button, also for a cancelled or failed simulation that has them
- the results panel follows GET /events/simulation while it shows partial results; a new partial or the
  final results replace the shown status, and EstimatorsSelect loads the estimator on screen again
- EstimatorsSelect is the only loader of estimator pages: it remembers the partial revision each tab was
  loaded at, loads the tab on screen (not always tab 0) and does not switch to the Results screen
- estimator pages are cached per partial revision, so a new partial is fetched and the final results
  never come from a cached partial

Co-authored-by: Mosder <xmosder12@gmail.com>

* feat: follow the status of all simulations through one stream per user, in place of polling

The status is polled only when the stream is refused three times in a row; a failed or canceled run reaches the
results panel too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix: open another simulation at its first estimator and keep only the latest partial pages

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix: keep trying the status stream while polling, without its stale statuses, and give another user a stream of their own

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…2442)

The status stream re-renders the provider on every pushed status; the handlers were bound anew each render,
so every grid and effect depending on them fetched again, several requests a second per open page.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…#2443)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…can be opened again (#2445)

The grids saw the tracked run's final status again on every remount and stream reconnect and each time jumped
to the results.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

#2446)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

…ts for it (#2448)

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>