Validated improvement ready
Reviewed the mirrored spring-projectsgh-37159 busy-spin/interrupt fix in both near-duplicate SubscriberInputStream classes (spring-core, spring-web). The PR's own production edits were already minimal and correct: the re-park condition and the required second volatile read in await() are both load-bearing for correctness (removing either reintroduces the busy-spin or a lost-wakeup race), and the duplicated read()/read(byte[]) catch blocks already existed pre-PR and extracting them would add control-flow indirection for two call sites with no measurable benefit, so no cleanup or performance edit was accepted. No build, dependency, or JDK path was touched by the PR, so modernization is not applicable. The one genuine gap found was test coverage: only spring-web's SubscriberInputStreamTests got a regression test for the interrupt-while-parked fix; spring-core's near-duplicate class (used internally by the public DataBufferUtils.subscriberInputStream API) had no test at all covering this scenario, so it was added as a focused characterization test proving the same observable contract (InterruptedIOException with restored interrupt status, no thread leak/spin) through the public API.