sidecar channels: don't try to clean up previous batch state if in provider mode

If a node is providing sidecar channels, it won't be involved in the actual channel opening process.
So in case of a batch rollback (new `PrepareMsg` without a finalize of the previous batch), there is nothing to clean up.
In fact, the cleanup attempt creates log entries with errors:

```
2021-12-14 18:15:28.432 [INF] RPCS: Received PrepareMsg for batch=020xxxx, num_orders=1
2021-12-14 18:15:28.439 [WRN] FNDG: Unable to unregister funding shim (pendingChanID=2dxxxx) for order=8ff7xxx
2021-12-14 18:15:28.444 [DBG] FNDG: Cleaning up incomplete/replaced pending channel in lnd was unsuccessful for order=8ff7xxxx (channel_point=d3c5faxxxx:0), assuming timeout when funding: rpc error: code = Unknown desc = unable to find closed channel summary
```

Trace the batch rollback handling around PrepareMsg and the provider-mode sidecar channel path, focusing on why funding-shim cleanup is attempted after a new PrepareMsg. Reproduce a rollback in provider mode and verify that no cleanup-error log is emitted when the node was not involved in channel opening.

Written by the indexing model from the issue text.