[parthod0x]

What

Adds test coverage for two AntPendingSolutions paths that currently have none. Test-only — no production code is touched.

Why

AntPendingSolutions is how a node queues its computors' solutions for publication, and it had a live bug recently (#1024, fixed by #1025). Two real paths were still untested:

1. The capacity fail-safe (droppedFull). When the pool is full and findFreeEntry() can reclaim no slot, add() must return false and bump droppedFull rather than overwrite a live entry. Nothing in the suite filled the pool, so this was never exercised.

2. The explicit retirements (markObsoleteGateRejected / markObsoleteParentGone). The publisher retires entries through these. The existing tests only reach OBSOLETE implicitly via selection-time expiry, so the public retire methods, their stat counters (obsoleteGateRejected / obsoleteParentGone), and the "OBSOLETE is not a duplicate" revival branch in add() reached from an explicit retire were all untested. That revival branch is the exact logic next to the #1025 fix, so a regression there would silently strand a resubmitted solution.

Tests added

Built on the file's existing freshPool() / key() / ref() helpers:

• FullPoolRefusesAndCounts — fills to CAPACITY, asserts the next add() is refused and droppedFull == 1.
• ExplicitRetireIsCountedAndAllowsRevival — markObsoleteGateRejected is counted, stops selection, and a corrected resubmit on the same key revives the entry in place.
• ParentGoneRetireIsCountedSeparately — markObsoleteParentGone increments its own counter, distinct from the gate-rejected and expiry counters.

Verification

Built and run on x86_64-pc-windows-msvc (v143 toolset, /std:c++20, matching the test project): all 14 TestAntColonyPending tests pass — the 11 existing ones plus these 3. No K12-collision construction is needed, so this deliberately avoids the known-hard-to-test slot-recompute edge from #1024.

[cyber-pc]

@parthod0x LGTM, but you need change the target branch from main to develop.

[parthod0x]

Done — retargeted to develop. I also rebased onto develop so it stays a single clean commit (just the test; no release commit in the diff). Thanks for the review!