Use subEncoder for files sub

[HDegroote]

Original here: https://github.com/holepunchto/hyperdrive-next/pull/57

Marked as draft because I'm not sure exactly which parts of the code related to batches were work-arounds and which were on purpose. Note in particular that I am now passing the blobs through as a parameter instead of loading it again from this.db in a batch, which enables me to use the _db arg to pass the batch and to get rid of the _files arg (I couldn't figure out why a batch needed to reload the blobs, but there might have been a reason).

There's also a TODO marked in the PR to verify something in the close logic, where we recently made a fix with comment 'workaround to flush the batches for now.' => I think that's no longer relevant, but sanity checking that before fully removing

[HDegroote]

Also pushed another commit to refactor watch(), so it also uses FILES_SUB instead of defining its own subEnc

[HDegroote]

This PR has a bug when creating a batch on a non-ready core--looking into it.

[HDegroote]

Fixed the bug in a not-so-elegant way, by always reloading the blobs (as before this PR)

The current flow for a batch is:

• Pass the original db, as well as the newly created batch to the new Hyperdrive instance
• in drive._open:
  • Load the blobs core, using the original db (needed to access the header)
  • Set this.db to the passed-in batch, so all operations on the db now operate on the batch

  • (it's this last step which is pretty ugly, but not sure if to try to clean up the entire flow in this PR, or to leave that for another one)

[LuKks]

@HDegroote Because a batch instance doesn't have getHeader that's the issue, right? The easy fix is to hack it around and use batch.tree property when it's a batch, or to add the method to the Batch class

[LuKks]

I moved the branch to the repo itself while preserving your commits so I could do some debugging: https://github.com/holepunchto/hyperdrive/pull/344

[mafintosh]

landed in main