[darion-yaphet]

What problem does this PR solve?

Issue Number: Related to #3275

Problem Summary:

When max_retry=0, or earlier retries have exhausted the budget, the backup timer can terminate an in-flight RPC with EBACKUPREQUEST before its overall timeout.

For example, an RPC configured with a 10ms timeout and a 2ms backup threshold may fail at approximately 2ms even though the original request could still succeed.

What is changed and the side effects?

Changed:

• Skip the backup request when no retry budget remains, while keeping the current request in flight.
• Restore the original overall deadline before returning from the backup event handler.
• Add regression coverage for zero and exhausted retry budgets, naming-service and direct channels, synchronous and asynchronous calls, and finite and infinite timeouts.
• Document the behavior in both client guides.

Side effects:

• Performance effects: No additional request is sent when the retry budget is exhausted. Performance benchmarks were not run.
• Breaking backward compatibility: No public API or ABI changes. Calls that previously failed prematurely now wait for completion or the overall timeout.

---

Check List:

• Built brpc_channel_unittest successfully.
• Passed all 89 tests in brpc_channel_unittest on macOS.
• Verified the new regression test fails before the fix and passes afterward across 24 configuration combinations.
• Updated the Chinese and English client guides.
• Please follow Contributor Covenant Code of Conduct.