net: --connect peers are manual, as in Core

[bkeroack]

Fixes #819.

Problem

getpeerinfo reports a peer dialled for --connect as outbound-full-relay. Bitcoin Core opens every -connect address as ConnectionType::MANUAL and reports it as manual.

Fix

Both dial paths for --connect targets now use PeerConnType::Manual:

• the tip-follow dial of the pinned targets in run.rs, through a new P2PNode::follow_from_net_as(peer, typ) (follow_from_net keeps OutboundFullRelay for its other callers);
• the redial of remembered --connect hostnames in PeerHub::redial_remembered_with. addnode add and --connect hosts now redial the same way, so remembered_redials drops the per-host type.

The connection type feeds more than the label. For a --connect peer:

Reader	Before	After	Core
is_preferred_download (whether a stalled headers-sync peer can be replaced)	counted, as outbound-full-relay	counted: Manual added	fPreferredDownload counts manual peers
--seednode addr-fetch: the startup dial (empty addrman) and the 10 s fallback (fires below 2 live outbound-full-relay peers)	the fallback was held off by 2+ live --connect peers; the startup dial ran whenever addrman was empty	neither runs under --connect	never dials seednodes under -connect
expect_services_from_conn (NODE_NETWORK required)	required	not required	ExpectServicesFromConn is false for MANUAL

The stale-tip rotation already skips --connect, so the outbound_full_relay_* helpers it uses are unaffected.

Adding Manual to is_preferred_download also covers the second case in #819: a node whose outbound peers all come from addnode could never replace a stalling headers-sync peer.

Tests

• node_run_p2p_short (product run_p2p --connect) asserted connection_type == "outbound-full-relay". It now asserts "manual".
• redial_same_endpoint_in_addnode_and_connect_dials_once: a --connect-only host must redial as Manual.
• session_heartbeat_replaces_stalled_headers_sync_peer_only_with_another_preferred (renamed from ..._keeps_sole_preferred_...): the sole preferred sync peer is still kept; once a Manual peer is live, the stalled sync peer is replaced.
• follow_dial_targets_connect_bypasses_diversity asserts follow_dial_type. New seednodes_are_off_under_connect covers seednodes_allowed, the guard both seednode paths share.

Each new assertion fails when its guard is reverted.

Ran locally: cargo fmt --check, ./scripts/ast-grep.sh, cargo clippy --workspace --all-targets -- -D warnings, the rbitcoin-net peers:: tests, the rbitcoin-node run:: tests, and node_run_p2p_short.

Not changed

is_preferred_download still leaves out Core's other preferred cases (feelers, inbound peers with noban), and it does not check that the peer can serve blocks.

Literal --connect addresses (IP, onion, I2P, CJDNS) are dialled once at startup and not redialled after the session drops. Only --connect hostnames go into the 2 s redial loop (set_connect_hosts(config.listen.connect_dns…) in run.rs). That is separate from #819 and unchanged here.

🤖 Generated with Claude Code