fix: respect reconnect=false and clamp server-supplied retry: 0

[kinyoklion]

Summary

Stacked on top of #134. Addresses three pre-existing reconnect-control gaps surfaced during the multi-agent review of that PR.

1. retry: 0 collapses backoff to zero. A server emitting retry: 0 set BackoffRetry::base_delay to Duration::ZERO, after which every reconnect path (client.rs:474, 525, 547, 612 and the new parse-error path) computed next_delay() == 0 and reconnected immediately — a tight loop. Clamps change_base_delay to a 1 ms floor.

2. EOF arm ignored reconnect_opts.reconnect. The body-exhausted branch unconditionally scheduled WaitingToReconnect even when reconnect was disabled. Now honors the flag and transitions to StreamClosed, matching every other error path.

3. Parse-error arm with reconnect=false left the parser poisoned. No state transition happened, so the next poll drained the broken body to EOF — where (2) above papered over the bug. Now transitions to StreamClosed so the documented "do not use the stream after error" contract holds.

Context

Stacked PR — base is the rl/sdk-2345/parser-error-reconnect-state branch from #134, not main. Will retarget once #134 lands.

Tracked in SDK-2347. Predecessor: SDK-2345 / #134.

Surfaced from the multi-agent review of #134 (findings 1, 2, and the corresponding suggested follow-ups 4 and 5). All three were pre-existing — #134 only made the parse-error path more visible.

Test plan

• test_change_base_delay_clamps_to_minimum (retry.rs) — pins the 1 ms floor against change_base_delay(Duration::ZERO) and a sub-floor value.
• parser_error_closes_stream_when_reconnect_disabled (client.rs) — asserts the stream emits one InvalidLine then None when reconnect is off.
• eof_closes_stream_when_reconnect_disabled (client.rs) — asserts the stream emits the event, Eof, then None when reconnect is off.
• Existing parser_error_schedules_reconnect_immediately from fix: schedule reconnect after parse error during streaming #134 still passes.
• cargo test — 63 lib tests + 1 doc test pass.
• cargo fmt --check clean.

---

Note

Medium Risk
Changes the SSE streaming state machine and retry backoff behavior, which can alter reconnect/termination semantics for consumers; scope is contained and covered by new tests.

Overview
Fixes reconnect disablement edge cases in the SSE client. When ReconnectOptions.reconnect is false, parse errors and clean EOF now transition the stream to StreamClosed so the next poll returns None instead of scheduling a reconnect or continuing with a poisoned parser.

Hardens retry backoff against server-supplied retry: 0. BackoffRetry::change_base_delay now clamps the base delay to a 1ms minimum to avoid zero-delay reconnect loops, with a new unit test and additional stream-level tests covering the new close-on-error/EOF behavior.

^{Reviewed by Cursor Bugbot for commit d33131a. Bugbot is set up for automated code reviews on this repo. Configure here.}

[kinyoklion]

We may want this larger.