Enhancement: SSE reconnection with exponential backoff

[jhodapp]

Problem

When the Rust backend restarts during a deploy, the SSE EventSource connection receives an HTTP 500 and enters the CLOSED state (readyState = 2). The current implementation in src/lib/hooks/use-sse-connection.ts closes the connection permanently and sets an error state — it never retries.

This means users lose real-time updates (action/agreement/goal change notifications) until they manually refresh the page.

Current behavior

In use-sse-connection.ts:

• readyState === EventSource.CONNECTING → browser auto-retries (network errors)
• readyState === EventSource.CLOSED → connection is closed permanently, no retry

Proposed solution

Add manual reconnection with exponential backoff when the EventSource enters the CLOSED state due to a server error (500/502/503). Key considerations:

1. Exponential backoff with jitter — base delay 1s, max 30s, add random jitter to prevent thundering herd when many clients reconnect after a server restart
2. Max retry limit — stop after ~8 attempts to avoid infinite retry loops
3. Reset on success — clear retry counter when connection is re-established
4. Don't retry on auth errors — 401/403 should still close permanently (the current force_logout SSE event handles this)
5. Consider migrating from native EventSource to fetch + ReadableStream — native EventSource cannot send custom headers (API version header), and the retry behavior is browser-dependent

Files to modify

• src/lib/hooks/use-sse-connection.ts — add retry logic
• src/lib/stores/sse-connection-store.ts — potentially track retry count/state

Research notes

• Native EventSource behavior on HTTP 500 varies by browser (spec says CLOSED, most browsers agree)
• @microsoft/fetch-event-source is the most popular library for controlled SSE (~1.5M weekly downloads) but hasn't been updated since 2023
• A manual fetch + ReadableStream implementation avoids the dependency and gives full control over headers and retry