Fix eventsource reconnection: proper backoff and close() behavior

[jvosloo]

Summary

Fixes three bugs in the EventSource plugin's reconnection logic that cause rapid-fire reconnection attempts when the server is unreachable, and prevent .close() from working in error handlers.

Bug 1: XOR instead of exponentiation

// Before (bitwise XOR — 2 ^ 2 = 0, giving 0ms delay\!)
var timeout = Math.random() * (2 ^ stub.retryCount) * 500;
// After (proper exponential backoff)
var timeout = Math.random() * Math.pow(2, stub.retryCount) * 500;

Bug 2: Backoff only triggers on CLOSED state

The error handler only reconnected when readyState == EventSource.CLOSED, but browsers keep readyState = CONNECTING on connection-refused errors and auto-retry every ~3 seconds — completely bypassing the backoff. Fixed by closing the EventSource on error to stop the browser's native auto-reconnect, then managing retry ourselves with proper exponential backoff.

Bug 3: .close() does not stop reconnection (#259)

Calling .close() in a user's on error handler sets readyState to CLOSED, which immediately triggers the plugin's own error handler to call open() — making the problem worse (as reported in #259). Fixed by:

• Adding a closed flag set by .close() and checked before scheduling reconnection
• Tracking the reconnect timeout ID and clearing it on .close()
• Resetting closed on .open() so reopening after explicit close still works

Backoff behavior after fix

Retry	Max delay
1	~1s
2	~2s
3	~4s
4	~8s
5	~16s
6	~32s
7	~64s (cap)

Closes #259