fix(#850): EventSource reconnect storm + backoff timing

[Hollujay]

Closes #850

Summary

Fixes the EventSource reconnect storm in useStreamEvents — supersedes #996 with a scoped-down diff per @ogazboiz's review. This PR touches only the two files directly related to the hook fix:

• frontend/src/hooks/useStreamEvents.ts
• frontend/src/__tests__/useStreamEvents.test.tsx

No CI workflow edits, formatting changes, logger/dashboard changes, or backend test tweaks are included — those were dropped from #996 as unrelated scope creep.

What changed

Reconnect storm fix:

• Memoized the subscription key (streamIds + subscribeToAll + jwtToken) so buildUrl doesn't get a new reference on every render, which previously caused the effect to reconnect even when the actual subscription hadn't changed.
• connect() now clears any pending reconnect timeout at the start, preventing overlapping/duplicate connections.
• Added a MAX_RECONNECT_ATTEMPTS cap (20) so a persistently failing connection stops retrying instead of looping forever.

Combined with an independent fix already on main:
While this branch was open, main picked up a separate fix for the backoff-delay timing (precomputing the capped delay and the next backoff value before scheduling setTimeout, rather than updating it after connectRef.current() already ran). Both fixes touch the same connect()/onerror block, so this PR merges them together rather than one overwriting the other — see useStreamEvents.ts for the combined logic.

Testing

• useStreamEvents.test.tsx now includes all 14 relevant tests: the original hook tests, main's 3 backoff-timing tests (exponential growth, cap at maxRetryDelay, reset after successful reconnect), and this PR's 2 storm-prevention tests (single EventSource instance across re-renders, stops reconnecting after hitting the cap).
• All 14 tests pass locally.
• Verified via tsc --noEmit that this change introduces zero new type errors compared to a clean main checkout.

Notes for reviewers

• The useIncomingStreams.test.ts → .tsx rename from the original [#850] Fix EventSource reconnect storm/leak in useStreamEvents #996 is not included here — it already landed on main via another PR, so re-including it would have re-created the collision.
• Happy to close [#850] Fix EventSource reconnect storm/leak in useStreamEvents #996 in favor of this PR once this is reviewed, or merge this into [#850] Fix EventSource reconnect storm/leak in useStreamEvents #996's branch instead if preferred — let me know what's easier on your end.