Share projection metadata read across factory instances

[zanarellidev]

• You have read the Spring Data contribution guidelines.
• You use the code formatters provided here and have them applied to your changes. Don't submit any formatting related changes. (No automated formatter goal is wired into mvn test/mvn package; matched the surrounding file's existing style by hand instead of running the IDE formatter directly.)
• You submit test cases (unit or integration tests) that back your changes.
• You added yourself as author in the headers of the classes you touched. Amend the date range in the Apache license header if needed. For new types, add the license header (copy from another file and set the current year only).

Closes #3513.

DefaultProjectionInformation.PropertyDescriptorSource.getMetadata() performed a fresh ASM
MetadataReader classpath read of a projection interface's .class file on every construction,
and ProxyProjectionFactory.projectionInformationCache is per-instance. Two independent
ProxyProjectionFactory/SpelAwareProxyProjectionFactory instances resolving the same projection
type each paid their own class-file read, with no sharing. In reactive applications this is the
blocking-I/O-on-event-loop Blockhound violation described in the issue.

This PR adds a shared, type-keyed ConcurrentReferenceHashMap (weak keys) cache for the metadata
read, below the per-factory layer, matching the approach the reporter proposed in #3513.

ProjectionMetadataCachingProbeTests reproduces the issue without needing R2DBC/Blockhound: it
uses a custom child-first ClassLoader that counts real .class resource reads, and confirms two
independent factory instances resolving the same projection type only trigger one read after this
fix (previously two).