Fix Int32 overflow in SSE retry field

[hahwul]

What

EventStream#send rendered the SSE retry field with .to_i (Int32). A retry span longer than ~24.8 days (Int32::MAX milliseconds) overflowed and raised OverflowError inside the streaming handler, breaking the response. Switched to .to_i64.

Bug case (before this PR)

sse "/events" do |stream, _|
  stream.send("update", retry: 60.days)  # 5_184_000_000 ms
  # => raises OverflowError: Arithmetic overflow
end

Values within the normal range (seconds/minutes) are unaffected — the emitted bytes are identical.

Test

Added a regression spec asserting retry: 60.days renders retry: 5184000000.

[hahwul]

It's a simple patch for overflow prevention. and CI passed, so I'll self-merge it!

[Sija]

Maybe using .to_u64 would be even better?

[hahwul]

Hi @Sija

Thanks for the suggestion! Nothing guards against negative spans here, so .to_u64 would raise on them inside the streaming handler. My reading was that negatives are harmless anyway since clients ignore the field per spec, hence .to_i64.

Happy to add a guard in a follow-up PR that skips the field when it's negative. With that in place .to_u64 works fine too, if you'd rather have the type say it. WDYT?