Drop the -blocksonly exclusion so recon outbound slots are reserved under blocksonly

This description is AI generated, mistakes can be made.

Intent: Make a -blocksonly -txreconciliation node allocate 4 OUTBOUND_FULL_RECONCILIATION slots. It never offers SENDTXRCNCL (net_processing checks ignore_incoming_txs), so every recon peer fails negotiation and gets disconnected. The slots churn forever and also shrink the inbound budget.

Behaviour change: Original: m_max_outbound_full_recon=0 under blocksonly, so ThreadOpenConnections never picks the recon type and addconnection outbound-full-recon is rejected. Mutant: max=4, the node keeps opening recon connections that get disconnected after the handshake, and m_max_automatic_outbound is 4 higher (4 fewer inbound slots).

Trigger: Start bitcoind with -txreconciliation -blocksonly and peers in addrman (or call addconnection outbound-full-recon on regtest, which now succeeds instead of failing).

Spec violation: PR's own comment: 'under -blocksonly these connections could never negotiate it and would waste their outbound slots'; BIP 330 negotiation is only offered by tx-relaying nodes.

Severity if undetected: medium

Why tests miss it: grep blocksonly in /work/pr-tests.diff and test/functional/p2p_sendtxrcncl.py (lines 207-210: inbound peer only), p2p_txrecon_*.py extra_args, denialofservice_tests.cpp (sets options directly)

Plausibility: Reads like a simplification: the -txreconciliation flag looks like it is enough on its own, and it is easy to assume blocksonly is already handled in net_processing.

Equivalence risk: If no test or user runs blocksonly with txreconciliation, nothing differs. But it is reachable by configuration, and the reconnect churn plus lost inbound slots are observable.