Improve sample generation performance by caching reflective type metadata

[SvenWirz]

Summary

This PR removes repeated reflection work from the sample generation hot path:

• Cached type metadata: introspected properties, accessible accessor methods and the public default constructor are now cached per type and CollectionSamplingMode in the new package-private TypeModel (see Properties.getTypeModel). Previously, Sample.newInstance() re-ran introspection and property filtering up to six times per generated instance - and recursively for every auto-sampled object in the graph.
• Cached generic type resolution: ReflectionUtil.getCollectionType caches the resolved collection element type per read method instead of resolving getGenericReturnType() on every call.
• Constructor reuse: auto-sampling reuses the cached default constructor instead of probing type.getConstructor() twice per generated object, and setAccessible(true) is no longer invoked on every single property write.

Behavioral notes:

• Properties.getProperties now returns an unmodifiable set; the only mutating caller (Sample.getNotHitFields) copies it first.
• Exception semantics are preserved (missing default constructor still raises SampleException / ReflectionException as before).
• The caches are static and never evict - the same trade-off as the existing InvocationSensor proxy class cache.
• A dedicated JaCoCo rule was added for TypeModel, following the existing pattern for internal classes with defensive code paths.

Benchmark

Adds a JMH benchmark (SampleGenerationBenchmark, test scope) covering flat bean generation, auto-sampling of an object graph (nested beans, collections, enum, map) and the configuration API including the field selector. Run it with:

mvn test-compile org.codehaus.mojo:exec-maven-plugin:3.1.0:exec -Dexec.executable=java "-Dexec.args=-classpath %classpath com.remondis.resample.benchmark.SampleGenerationBenchmark" -Dexec.classpathScope=test

Measured before/after on this branch (JDK 21, average time, 1 fork, 3 warmup + 5 measurement iterations):

Benchmark	Before (us/op)	After (us/op)	Change
flatBean	1.483 +/- 0.026	1.235 +/- 0.029	-17%
objectGraphWithAutoSampling	16.293 +/- 0.163	11.688 +/- 0.199	-28%
configureAndCreateFlatBean	2.513 +/- 0.060	1.874 +/- 0.072	-25%

All 50 runnable unit tests pass; formatter validation and the JaCoCo coverage rules pass as well.

🤖 Generated with Claude Code