Reconciliation slots left out of m_max_automatic_outbound

This description is AI generated, mistakes can be made.

Intent: Recon slots are not reserved: semOutbound is sized 4 too small and m_max_inbound 4 too large, so total connections can exceed -maxconnections (file-descriptor budget) and recon peers compete with full-relay/block-relay/feelers for outbound grants.

Behaviour change: With -txreconciliation and default -maxconnections=125: original reserves 15 outbound grants and 110 inbound; mutant reserves 11 outbound grants and 114 inbound, so with 8 full-relay+2 block-relay+1 feeler outstanding no recon connection can acquire a grant, and fully saturated inbound plus recon peers exceed maxconnections by 4.

Trigger: Start bitcoind with -txreconciliation on mainnet; ThreadOpenConnections fills 8 full-relay and 2 block-relay slots, then a feeler grant is held while trying to open recon peers; also an inbound flood to the inbound limit.

Spec violation: PR comment in Init: 'reconciliation slots are subtracted here like any other outbound, shrinking our inbound budget.'

Severity if undetected: medium

Why tests miss it: p2p_connection_limits.py (runs without -txreconciliation, so recon budget is 0 and identical), p2p_reqtxrcncl.py / p2p_txrecon_initiator.py open at most a handful of addconnection peers (well under 11 grants), denialofservice_tests uses AddTestNode which bypasses semOutbound

Plausibility: It is the pre-PR line verbatim; forgetting to add the new budget to the sum is a classic omission when adding a new connection type.

Equivalence risk: When -txreconciliation is off m_max_outbound_full_recon is 0 and the mutant is equivalent; it differs only with reconciliation enabled.