We previously allowed users to disable anchor channels and drain their
anchor reserve while still having anchor channels open or pending
resolution. This was acceptable for keyed anchor channels, as the
commitment transaction therein still contained some fees, and had some
chance of getting mined into a block without any anchor bumps.

In upcoming commits, we will add support for 0FC channels, and their
commitment transactions have zero fees and depend entirely on the anchor
reserve to reach miners and get confirmed in a block. It is thus
dangerous to disable anchor channels and drain the reserve after 0FC
channels have been opened. Therefore, we make `AnchorChannelsConfig`
required, and prevent this case from ever happening.

The patch adds support for the `broadcast_package` method added in
electrum protocol v1.6. Upcoming commits will require this patch to pass
CI.

The mempool/electrs docker image used in those tests only supports
submitpackage via the esplora interface, not the electrum interface.

We bump the Bitcoin Core version used in kotlin and python tests to
support ephemeral dust. This is required for 0FC channels.

In upcoming commits we will read this knob to determine whether to
negotiate 0FC channels.

For now, we make a best-effort attempt to make sure the configured chain
source supports 0FC channels if this knob is set.

Do this roundtrip at the same time we make a roundtrip to retrieve the
feerates to keep startup as fast as possible.

Implementations of `BroadcasterInterface` cannot assume any topological
ordering on the transactions received, so here we order the received
transactions before adding them to the broadcast queue. Any consumers of
the queue can now assume all transactions received to be topologically
sorted.

Codex wrote the tests.

These will be useful when we add support for broadcasting packages in an
upcoming commit.

We rely on the `BroadcasterInterface` contract whereby any
multi-transaction vector must be a single child and its parents.

In a prior commit, we added the guarantee that any packages received
from the broadcast queue are already topologically sorted, and hence
can be passed directly to the `submit_package` Bitcoin Core RPC.

We avoid broadcasting non-TRUC parents-child packages via
`submitpackage` for now to avoid adding a requirement to support
`submitpackage` for users that don't enable zero-fee commitment
channels. We will do so once support for `submitpackage` is more
ubiquitous.

The electrsd dev-dependency does not yet ship binaries that support
submitpackage on electrum, so we do not turn on 0FC channels in local
tests, and instead only turn them on in a standalone CI workflow, which
includes a custom electrs server that adds support for submitpackage
on electrum.

Co-Authored-By: HAL 9000

Anchor reserve accounting only looked at channels returned by the
ChannelManager. That misses an important force-close transition:
ChannelManager::list_channels can stop returning a channel before
its commitment transaction confirms. During that window the
ChannelMonitor can still broadcast, rebroadcast, or fee-bump the
holder commitment and its anchor spend, so the wallet's emergency
reserve must remain available.

We now require a anchor reserve to be kept for non-trusted channels
where the type is not yet known; we've recently made anchor channels
always supported in ldk-node, so at this point it is safe to assume most
channels of ldk-node will be anchor channels.

We now no longer count existing 0FC channels to validate support
for submitpackage on the chain source if those channels are in the
trusted peers list.

Co-Authored-By: HAL 9000

Store-backed integration tests drive wallet syncs explicitly. Leaving
Esplora background sync enabled let the first background scan start
before the funded address was tracked, so a later sync_wallets call
could observe a zero premine balance.

Reuse the manual Esplora sync config from shared test setup for
Postgres, VSS, migration, and current-version compatibility builders.

Co-Authored-By: HAL 9000

Prober::locked_msat only counted pending probe records from
ChannelManager::list_recent_payments. Across restart, LDK can expose a
still-locked outbound HTLC through channel state before the matching
pending probe record is visible again, which made the restored locked
budget read as zero.

Fall back to unresolved outbound channel HTLCs when no recent pending
probe accounting is available. Also make the restart test retry the
probe-and-restart boundary if the probe failure wins the race before the
node state is persisted.

Co-Authored-By: HAL 9000