[AkashB23]

Fixes #3321

BindingServiceProperties.bindings became a plain TreeMap again in GH-3158, which dropped the thread safety added by #3011. The map is written on read (bindIfNecessary -> bindToDefault -> bindings.put) by StreamBridge, BindingService and the binders, from many threads and without a lock, so concurrent first use of binding names corrupts it (lost entries, size() not matching the iteration, a NullPointerException in TreeMap.rotateRight, or a thread spinning forever in TreeMap.successor). Details, the history of the field and the affected versions are in #3321.

This changes the map to new ConcurrentSkipListMap<>(String.CASE_INSENSITIVE_ORDER):

• it is safe for concurrent use;
• it keeps what Spring Cloud Stream binding names in properties have become case sensitive #3158 needs: case-insensitive lookup, the original key casing and sorted iteration, as with the TreeMap;
• a ConcurrentHashMap is not an option, since it is case-sensitive (the reason for Spring Cloud Stream binding names in properties have become case sensitive #3158).

#3076 replaced the ConcurrentSkipListMap for performance, but the reporter saw no improvement with the ConcurrentHashMap build and traced the slowdown to a Spring Boot child-context change and a misplaced output-bindings property, so the map was not shown to be the cause (see #3321 for the details and a measurement of the per-lookup cost).

Tests

BindingServiceTests#concurrentFirstUseOfBindingsKeepsTheMapConsistent (new): 16 threads call getBindingProperties for the same 1000 new names in different orders, for 5 rounds, and the test checks that nothing hangs or throws and that no entry is lost.

I ran this method in two batches of 30 against the unmodified class, where it failed 29 and 27 times (mostly NullPointerExceptions from the corrupted tree, otherwise lost entries or a thread that never finished), and 30 times against the patched class, where it passed every time. I did that with the 5.0.2 jars outside Maven, because I could not run the Maven build of main locally (its snapshot parent POM was not resolvable for me). So I have not run the rest of the module's tests, including ExplicitBindingTests, which GH-3158 touched; I am relying on CI for those.

[AkashB23]

@olegz can you please check