Previously reading from an empty, closed Pipe returned:
* IO exception if there were no writes to the pipe,
* Empty byte array if there were writes to the pipe.
With this change, we return an empty byte array on both paths.
This also fixes a subtle race condition in the following flow:
1. Pipe is empty
2. Thread A is reading from the pipe (blocking call).
3. Thread B writes to the pipe and immediately closes it.
4. Thread A is unblocked and returns IO error. The data written by thread B is
ignored!
The new test in bluetooth_classic_test triggered that defect consistently.
PiperOrigin-RevId: 518107700
Guogang Li