chantools forceclose --channeldb crashes on taproot channels: malformed signature: no header magic

Nobody has claimed this yet.

## Problem

I stopped LND and ran the following command:

```
$ chantools --testnet forceclose --channeldb ./.lnd/data/graph/testnet/channel.db --fromchanneldb ./.lnd/data/graph/testnet/channel.db
Input your 24-word mnemonic separated by spaces: ***
Input your cipher seed passphrase (press enter if your seed doesn't have a passphrase):
```

If the node does not have a taproot channel, this command completes successfully:

```
2023-12-31 18:32:20.935 [INF] CHAN: Writing result to results/forceclose-2023-12-31-18-32-20.json
```

After I opened a taproot channel, it started crashing with the following error:

```
Error: malformed signature: no header magic
Usage:
chantools forceclose [flags]
...
```

I verified this observation on two nodes (testnet and signet).

## Analysis

I think, that the root cause is in the function [LightningChannel.CreateSignDesc](https://github.com/lightninglabs/chantools/blob/5cf7fd60c48174608f0330d08298b35472425a17/lnd/channel.go#L24-L54) which is not aware of taproot channels.

lnd's function [createSignDesc](https://github.com/lightningnetwork/lnd/blob/4d8fa349ca405a42548717c17cc689e6612ff37a/lnwallet/channel.go#L1492) has taproot aware implementation.

In https://github.com/lightninglabs/chantools/pull/95 I'm adding the code building SignDescriptor from a channel backup. I think most of that code can be factored into a generic function, which can be used by LightningChannel.CreateSignDesc as well. Maybe even chantools can use a common function from LND to avoid diversions in the future.

Start with LightningChannel.CreateSignDesc in lnd/channel.go and compare it with LND's taproot-aware createSignDesc implementation. Review the SignDescriptor construction discussed in pull request 95, then reproduce the forceclose command with a taproot channel. Done means taproot forceclose no longer reports a malformed signature while non-taproot channels continue to work.

Written by the indexing model from the issue text.