Fix resource safety issues in topic message decoder

[Myllyenko]

1. Guard the ready handler and validate the buffer size

Change	Old behavior	New behavior
Buffer configuration validation	A zero or negative maxBufferSize produced a decoder that can never admit a message, because DecodeNext only runs while totalAvailable is positive. The reader stalled silently.	Construction fails immediately with IllegalArgumentException.
Ready-handler failure	readyHandler was invoked directly from EncodedMessage.decode() and from setError(). An exception thrown by the handler escaped into the decompression executor thread, where it is reported as an uncaught exception and can suppress decoding of the messages that follow.	The handler is invoked through notifyReady(), which logs the failure after the message has already been marked as ready.

This commit changes no existing expectations — all pre-existing tests pass untouched.

2. Make the decompression buffer limit a hard limit

Change	Old behavior	New behavior
Buffer-limit enforcement	DecodeNext admitted the next message whenever any budget remained, without looking at its size, so totalAvailable went negative and the decoder retained more uncompressed data than maxMemoryUsageBytes allows.	A message is admitted only when it fits into the remaining budget. A message larger than the whole budget is still admitted, but only once nothing else retains the buffer, so the overshoot stays bounded by that single message.

This commit is up for discussion and can be dropped on its own. Four existing tests (flowControlByBudgetTest, partitionFlowTest, decodeStopTest, decodesOnProvidedExecutorTest) asserted the previous behavior explicitly through negative getTotalAvailable() values, which suggests the overshoot may be an intentional trade-off rather than an oversight. Their expectations are updated in this commit. The trade-off of the stricter rule: a message larger than the budget now waits until the buffer is fully released, and since one decoder is shared by all partitions of a reader, that wait can be longer when several partitions are active.

If the greedy admission is intended, dropping commit 2 leaves commit 1 applicable as is.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 73.22%. Comparing base (d77889d) to head (aa48da9).

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #710      +/-   ##
============================================
+ Coverage     73.18%   73.22%   +0.03%     
- Complexity     3589     3592       +3     
============================================
  Files           392      392              
  Lines         16531    16546      +15     
  Branches       1736     1739       +3     
============================================
+ Hits          12099    12116      +17     
  Misses         3811     3811              
+ Partials        621      619       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.