…op (#602 follow-on)

Wires the pure ratchet dgb::ratchet_min_protocol_version (#602) into the node
runtime, the SUT equivalent of oracle p2pool-dgb-scrypt main.py:213-216 +
data.py:857 update_min_protocol_version. On each best-share advance (the three
m_best_share_hash update sites under m_tracker_mutex) NodeImpl::apply_min_protocol_ratchet
samples the work-weighted desired-version tally over the SAME window the 60%
version-switch gate reads (share_check step 2: [CHAIN_LENGTH*9/10, CHAIN_LENGTH]
behind the best share PARENT) and lifts the inbound P2P accept-floor 1400 -> 3500
once the best share VERSION holds >=95% of that work.

The floor moves from the immutable config value to a latching atomic
m_runtime_min_protocol_version (seeded from PoolConfig::MINIMUM_PROTOCOL_VERSION);
handle_version reads it lock-free, the compute thread publishes via store. Adds
the call-site guard the pure ratchet omits: apply_min_protocol_ratchet_decision
enforces oracle main.py:212 (len(shares) > CHAIN_LENGTH) so a fresh node with a
partial/empty window never spuriously locks the floor to 3500 and rejects every
peer -- the pure fn ratchets on an empty window (dict-branch), the wired decision
must not.

KAT dgb_min_protocol_ratchet_test +5 wiring cases (14 total, all green): short-chain
never-lifts (vs pure-fn ratchets on same weights), short-chain empty-window no-lift,
full-window unanimous lifts, full-window below-95 stays cold, already-at-target
short-circuits. Fenced to src/impl/dgb/; c2pool-dgb links clean (incl -DAUX_DOGE=ON).