[Draft] Fix flushing of rejects before disconnecting

[Liquid369]

Dash dashpay#3429

[mkorovkin2]

NACK

Reviewed and tested merge result against current master.

Blockers:

• DisconnectNodes neither advances it for connected peers nor preserves it after erase, causing an infinite loop/undefined behavior while holding cs_vNodes.
• GDB attached to the exact test-created pivxd: its hot b-pivx-net thread was stopped in this loop while b-pivx-msghand waited on cs_vNodes; an iterator-only fix made the same cache workload finish in 14.86s.
• The claimed reject-flush purpose is not implemented: reject state/consumer were restored, but nothing enqueues CBlockReject; this also partially reverses PIVX's deliberate BIP61 removal without rationale or tests.

Additional issues:

• The 100ms linger busy-spins under cs_vNodes, and the partial one-second cleanup backport breaks disconnect, isolation, and reconnect timing.
• Inactivity checks are duplicated; byte accounting is dead; teardown logs use global rather than peer masternode state; no regression tests were added.

Validation:

• Exact PR: clean Linux build and full make check passed, but both 105-test default and 108-test extended runners failed during mandatory cache creation; fixed master completed it in 15.68s.
• Disposable runtime correction: full rebuild and make check passed; default finished 101 pass/4 skip/0 fail, extended 104/4/0, and supplementary ZMQ passed. Three Linux-only bind cases remain uncovered due Docker storage EIO.
• All 14 scripts excluded from the maintained runner were also executed separately: 2 passed and 12 failed from known legacy test drift.

Do not merge until iterator safety, bounded non-spinning cleanup, PIVX reject policy, and regression coverage are resolved. Lmk if there's anything I missed here as well.