lnsweep: anchor channels with several HTLCs are not handled correctly

[ecdsa]

If a channel has anchors, lnsweep creates a first-stage htlc tx for each htlc, which includes a UTXOs from the wallet. If the commitment tx has several htlcs, we currently reuse the same UTXO in all htlc transactions, because it is still 'unspent' when we loop over htlcs. This can be tested in various regtests by forcing the client to perform a MPP:

+    $alice setconfig test_force_disable_mpp false
+    $alice setconfig test_force_mpp true

Instead of calling wallet methods from lnsweep, we should delegate the creation of the htlc tx to the wallet.
This could be achieved by extending the "batch payment manager" in #9298.

Instead of being payment driven, the payment manager should accept 3 types of requests:

• include a given output (for payment requests)
• include a given input (sweep to the wallet). (we need that in order to CPFP anchor outputs)
• include a given input and a given output (we need that in order to include zero-fee first-stage htlc transactions)