feat(bootstrap): dial always-on bootstrap peers so a fresh node has a dialable peer

[MichaelTaylor3d]

DO NOT MERGE — gate round not yet run.

Closes nothing on its own; refs https://github.com/DIG-Network/dig_ecosystem/issues/923.

What changed

Adds the bootstrap seam (crates/dig-node-core/src/seams/dig_peer/bootstrap.rs) and wires it into
peer-network bring-up: resolve the anchor set, dial each over the full NAT traversal ladder, adopt
each verified connection into the connected-peer pool.

A freshly-installed node knows no peers. Peer exchange spreads only the peers a link's far end
knows, the DHT routes queries through peers already in the table, and a relay reservation makes the
node reachable without ever populating an address book. So a node on a machine that has never run
one has nothing to dial and reports connected_peers = 0 indefinitely.

Two findings that shaped the design

1. An address alone is not dialable. GossipHandle::connect_via_nat pins the peer's certificate
   SPKI against an EXPECTED peer_id, and dig-node's existing PEX dial loop (pex.rs:613) already
   continues on any candidate lacking one. So a bootstrap entry needs an identity, not just an
   authority. An entry without one is SKIPPED rather than dialed unpinned — an unpinned dial would
   accept whatever identity answered at that address.
2. dig-constants 0.11 cannot be adopted yet. The canonical set landed in dig-constants 0.11.0
   (feat(bootstrap): canonical DIG_BOOTSTRAP_PEERS so a fresh node has a dialable peer dig-constants#13), but this workspace pins dig-constants to 0.9: 0.10 moved to
   chia-protocol 0.36 while everything here builds against 0.26, and linking both stops Bytes32
   unifying across the workspace. That is the pin's own documented rationale
   (crates/dig-node-core/Cargo.toml:227), and I verified it empirically — bumping to 0.11 produces
   E0308 ... multiple different versions of crate chia_protocol. So compiled_in_targets() is
   explicitly empty and documented as such, and DIG_BOOTSTRAP_PEERS is the supported way to name
   an anchor until the chia 0.36 migration lands.

Neither of these is a shortcut: the anchor's own peer_id is not known yet either (see the report),
so the canonical entry would be skipped as unpinned regardless. The dial mechanism is complete and
exercised end-to-end through the env path.

Blast radius

impact was unavailable for this worktree (see report); radius established by grep + direct read
and stated honestly.

• bootstrap.rs is a new module with no existing callers.
• The one edit to existing code is an additive call in run_peer_network (peer.rs), plus a
  pub mod and a re-export line. No existing function signature, type, or test was modified.
• run_peer_network is the peer-network bring-up entry point, so it is high-traffic — which is why
  the added call is a spawn that cannot fail bring-up and returns early when the target set is
  empty (the current default). With no anchors configured the behaviour is byte-identical to before
  apart from one INFO log.
• No HIGH/CRITICAL risk to warn on: the change adds a code path and removes none.

How verified (locally — dig-node's Test + coverage check is known false-green, #2513)

• cargo test -p dig-node-core --lib — 854 passed, 0 failed.
• The 12 new bootstrap tests pass by name.
• cargo clippy -p dig-node-core --all-targets -- -D warnings — clean, exit 0.
• cargo fmt --check — clean.

Test design notes

Fixtures were built to distinguish the property from the nearest wrong implementation, not merely to
assert an outcome:

• env_list_parses_every_entry uses two entries — one cannot tell "parses the list" from
  "parses the first element".
• a_malformed_entry_is_dropped_without_discarding_the_list pairs each bad entry with a good
  neighbour — a list of only-bad entries yields empty either way and cannot tell "drops the entry"
  from "rejects the list".
• a_blank_value_falls_back_rather_than_disabling carries a populated control through the same
  argument position, so it cannot pass merely because both arms are empty today.
• compiled_targets_are_absent_for_now_and_pinned_whenever_present asserts the current empty state
  AND the pinning invariant, so it begins checking pinning rather than going vacuous the day the
  canonical set is adopted.
• authority_resolution_prefers_ipv6 guards its assertion on the host actually offering both
  families, so it cannot pass vacuously on a single-family host.

Version

Workspace 0.126.0 → 0.127.0; dig-node-core 0.47.0 → 0.48.0 (minor: additive capability).

Not in this PR

dig_ecosystem#1871 (adopt_relayed_inbound never called) is not fixed here — it is blocked on a
dig-gossip API gap, detailed in the lane report. Shipping the naive call would count the peer while
stopping its peer-RPC from being served, which is a regression, not a fix.