fix(consensus): keep a late retry's reply instead of refusing it

[numinnex]

No description provided.

[numinnex]

Merge after #3944

[codecov]

Codecov Report

❌ Patch coverage is 93.40659% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 68.20%. Comparing base (ffb186d) to head (28035ad).

Files with missing lines	Patch %	Lines
core/consensus/src/client_table.rs	93.25%	3 Missing and 3 partials ⚠️

Additional details and impacted files

@@              Coverage Diff              @@
##             master    #3945       +/-   ##
=============================================
- Coverage     84.20%   68.20%   -16.00%     
  Complexity     1398     1398               
=============================================
  Files          1219     1218        -1     
  Lines        173564   154236    -19328     
  Branches     140662   121335    -19327     
=============================================
- Hits         146153   105201    -40952     
- Misses        23449    45119    +21670     
+ Partials       3962     3916       -46     

Components	Coverage Δ
Rust Core	64.45% <93.40%> (-20.57%)	⬇️
Java SDK	67.23% <ø> (ø)
C# SDK	75.41% <ø> (+0.02%)	⬆️
Python SDK	90.06% <ø> (ø)
PHP SDK	85.65% <ø> (ø)
Node SDK	96.07% <ø> (ø)
Go SDK	69.09% <ø> (ø)

Files with missing lines	Coverage Δ
core/consensus/src/metadata_helpers.rs	94.03% <100.00%> (+0.01%)	⬆️
core/consensus/src/client_table.rs	90.06% <93.25%> (+0.11%)	⬆️

... and 305 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hubcio]

one more that lands on a line outside the diff, so leaving it here:

encode_decode_roundtrip_preserves_dedup_state (core/consensus/src/client_table.rs:3044) builds a 3-deep ring, so the new transferable_replies() cap is exercised by nothing - shipping 0, 5, or the whole ring would all pass. commit more than REPLY_RING_CAPACITY replies, encode/decode, then assert exactly the newest 5 replay and the 6th-newest draws AlreadyApplied. that's also the concrete test for the transfer guarantee question on line 1720.

[numinnex]

/ready