THRIFT-6125: Close Ruby client transports after uncertain sends

[kpumuk]

Ruby clients currently close their output transport when argument serialization fails, but leave it reusable when writing the message envelope, finishing the message, or flushing fails. At that point, request delivery may be partial or otherwise uncertain, so a subsequent call can reuse transport state that no longer has a reliable request boundary.

This change prepares generated arguments before consuming a sequence ID or writing message bytes. Once message output begins, envelope writing, argument serialization, message completion, and flushing share one failure boundary that closes the output transport while preserving the original exception. Ordinary calls are added to the pending-reply queue only after a successful flush, and oneway calls follow the same transport-closing behavior.

• Did you create an Apache Jira ticket? THRIFT-6125
• If a ticket exists: Does your pull request title follow the pattern "THRIFT-NNNN: describe my issue"?
• Did you squash your changes to a single commit? (not required, but preferred)
• Did you do your best to avoid breaking changes? If one was needed, did you label the Jira ticket with "Breaking-Change"?
• If your change does not involve any code, include [skip ci] anywhere in the commit message to free up build resources.