Replace the 0xsiddharthks/kyoto fork (siddharth/upgrades, rev f2ba601c)
with crates.io bip157 = "0.5.0". All four features the fork carried
landed upstream:

  - get_header(height)            -> PR #563
  - height_of_hash(hash)          -> PR #564 (canonical-only semantics)
  - whitelist_only()              -> PR #565 folded into #567
  - hostname trusted peers        -> PR #566 (redesigned in review)

Adapt to upstream's API changes:

  - DnsPeer removed. Use TrustedPeer::from_hostname(host, port) and the
    Builder::add_peer{,s} APIs over TrustedPeer instead of the old
    add_dns_peer{,s}.
  - Note v0.5.0 consumes hostname peers on use (popped from the
    whitelist when first tried, never reinstated). Re-resolution
    requires rebuilding the kyoto node, which our connectivity
    supervisor already does on consecutive failures.
  - broadcast_tx renamed to submit_package(impl Into<Package>). We
    pass a Transaction directly via the From<Transaction> for Package
    impl, so the call site is a one-line rename with the same return
    type (Wtxid).

Requester::height_of_hash now returns None for non-canonical hashes
(canonical-only via #564). The hashi call site already treats Ok(None)
as "block not on canonical chain" so this tightens correctness for
free with no behavior change. Soften the surrounding warn from
"possibly malicious" to acknowledge sync lag and recent reorgs as the
more common explanations.

We also inherit BIP-130 sendheaders, the new/tried peer fallback fix,
the late-CF-header ban fix, plus rescan_from(height), chain_tip(),
peer_info(), and Package::new_one_parent_one_child for future use.

bip157 v0.5.0 pops hostname trusted peers from the whitelist on use,
so a single peer drop ends the kyoto Node task with NoReachablePeers.
The event loop kept waiting on now-silent kyoto channels while
client_rx accepted deposits, so deposits stalled indefinitely (CI
test_large_withdrawal_signature_chunking confirmed only 27/500 over
600s; see kyoto Failed to look up block ... receiver was dropped).

Race the event loop against the kyoto JoinHandle so any exit (Ok,
NodeError, or panic) routes into the existing rebuild path.