fix: handle deleted segments gracefully instead of looping on 404

[davemun]

Ruby SDK

What did you accomplish?

When a segment is deleted in the Split UI, the SDK enters an infinite error loop because .segments.registered is append-only and the 404 response from segmentChanges raises an exception that gets retried every polling interval.

This PR handles 404 responses gracefully by returning nil from fetch_segment_changes, removing the segment from the registered set, and breaking the fetch loop.

How to test new changes?

bundle exec rspec spec/engine/api/segments_spec.rb

Extra Notes

We use split-proxy and we are seeing that every time we delete/rename a segment in Harness/Split, the various deployed Ruby Split clients will hammer split-proxy with requests for that deleted segment until that SDK is stopped and restarted.

This issue exists in your other implementations:

• Java
  • https://github.com/splitio/java-client/blob/master/client/src/main/java/io/split/client/HttpSegmentChangeFetcher.java#L67-L74
    • All non-2xx throw IllegalStateException, no 404 handling
• JS
  • https://github.com/splitio/javascript-commons/blob/development/src/services/splitHttpClient.ts#L60-L81
    • 404 gets a custom message string but still throws NetworkError, no unregistration
• Go
  • https://github.com/splitio/go-split-commons/blob/main/service/api/client.go#L86-L100
    • All non-2xx return generic HTTPError, no 404 handling
• Python
  • https://github.com/splitio/python-client/blob/master/splitio/api/segments.py#L61-L64
    • All non-2xx raise APIException, no 404 handling

[davemun]

afaik if this throws, the other segments that were supposed to get updated in this polling cycle also do not receive an update, which is a secondary issue of this problem