[Problem] Fix FlowableCache GC issue

[DragonFSKY]

Summary

• Move the source-facing cache state and subscriber registry into a static Multicaster, so the upstream connection and active CacheSubscriptions no longer retain the outer FlowableCache.
• Keep per-subscriber request accounting, cancellation, and serialized replay behavior unchanged.
• Retain the cache head only from FlowableCache, retain the current tail from Multicaster, and clear the tail on termination.
• Ignore post-terminal source signals now that termination clears the tail.
• Add a focused MemoryMXBean / System.gc() regression test matching the approach used for ObservableCache in Fix ObservableCache GC issue #7911.

Tests

• ./gradlew build --stacktrace
• ./gradlew javadoc --stacktrace
• FlowableCacheTest.valuesAreReclaimable repeated five times
• The regression test fails against the unmodified implementation because the cached payload remains retained.

References

• Addresses the FlowableCache item in 4.x: Cache/Replay operators to be more GC friendly #7912.
• Follows the ownership split used for ObservableCache in Fix ObservableCache GC issue #7911 while preserving Flowable-specific backpressure semantics.

Checklist

• The description explains what changed and why.
• The related issue and reference PR are included.
• A focused regression test is included; existing FlowableCacheTest coverage continues to exercise empty, range, error, backpressure, cancellation, and race behavior.

[github-actions]

🐷 TruffleHog + Entropy Beauty Scan

Average entropy of changed code: 4.647 bits/char
Verdict: ✅ Mid-4 beauty detected (thoughtful human code!)

Changed files entropy:

.gitattributes: 4.493
.github/ISSUE_TEMPLATE.md: 4.674
.github/PULL_REQUEST_TEMPLATE.md: 4.463
.github/dependabot.yml: 4.262
.github/workflows/compute-entropy.py: 4.850
.github/workflows/entropy-beauty-scan.yml: 4.583
.github/workflows/gradle-wrapper-validation.yml: 4.987
.github/workflows/gradle_branch.yml: 5.075
.github/workflows/gradle_jdk11.yml: 5.008
.github/workflows/gradle_pr.yml: 4.977
.github/workflows/gradle_release.yml: 5.190
.github/workflows/gradle_snapshot.yml: 5.018
.github/workflows/scorecard.yml: 4.752
.gitignore: 4.635
.gitpod.yml: 4.605
CONTRIBUTING.md: 4.704
COPYRIGHT: 4.927
DESIGN.md: 4.801
LICENSE: 4.427
README.md: 4.989

✅ No secrets or suspicious high-entropy strings found.

Mid-4 beauty heuristic in action — powered by our entropy chats! 😊

[akarnokd]

Looks okay, but how much do you need this in 3.x?

I'd prefer you primarily target 4.x first and then backport if someone does have a need for in 3.x.

I don't want to release more 3.x versions that aren't bug or crash fixes. This issue is more like GC pedantics than a grave mistake.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.61%. Comparing base (9b55d01) to head (b0c3676).
⚠️ Report is 234 commits behind head on 3.x.

Additional details and impacted files

@@             Coverage Diff              @@
##                3.x    #8247      +/-   ##
============================================
- Coverage     99.62%   99.61%   -0.01%     
+ Complexity     6801     6744      -57     
============================================
  Files           752      752              
  Lines         47707    47727      +20     
  Branches       6401     6404       +3     
============================================
+ Hits          47527    47544      +17     
+ Misses           84       80       -4     
- Partials         96      103       +7     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[akarnokd]

Also if you used AI, you have to disclose it, even for mechanistical changes.

I'm sorry but I have to do this on a high visibility attractive project like RxJava.

If you do not respond to change request in a reasonable time, indicated by your public github activity profile, or your contribution causes an effort amplification (i.e., I now have to port your changes over to 4.x and do all sorts of extra steps for no apparent reason or benefit), you will be banned and publicly shamed.

[akarnokd]

Please target 4.x first and we can discuss how much you, or anyone else needs this GC pedantic change backported into 3.x

[DragonFSKY]

Thank you for the guidance. I have opened #8249 as a clean 4.x version of this change, targeting the current 4.x branch as requested. I am closing this PR as superseded, and I will only revisit a 3.x backport if there is a concrete need after the 4.x change is reviewed.