Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Replace deprecated actions-rs/toolchain@v1 with dtolnay/rust-toolchain@stable
and the manual actions/cache@v3 setup with Swatinem/rust-cache@v2. The
purpose-built Rust cache action keys correctly on Cargo.lock + rustc version
(the old key only hashed Cargo.toml so cargo update wouldn't bust it), prunes
the parts of target/ that don't help on restore, and keeps build-script
outputs so the bitcoind/electrs binaries downloaded by corepc-node and
electrsd remain cached.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

build_test_nodes was funding the LSP, mining 6 blocks, then funding the
third-party node and mining another 6. The two fundings are independent —
broadcast both txs first and confirm with one batch of 6 blocks. Drops one
6-block confirmation wait (~10s on this machine) per integration test.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The split into 50+51 batches dates to a regtest setup pre-modern bitcoind
where 101 blocks could time out the default RPC. With current bitcoind 29.0
this fits comfortably in one call.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

The 1-second invoice has already expired by the time we need it expired;
sleeping 5x as long doesn't add safety, just CI time.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

create_lsp and create_third_party each called set_listening_addresses with
the same address twice. The second call replaces the first, so the second
call alone is equivalent.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>