…tening IDs

Co-authored-by: Eliah Kagan <degeneracypressure@gmail.com>

This allows to re-use that memory at least, making this kind of state a little more useful.
Also, these fields can certainly be considered stable.

That way, it's also possible to obtain the previous location in case of rewrites.

…es::Change` types

This is useful if in the middle of an edit you'd like to lookup what's there
to choose a non-conflicting name, for example.

…t passed on.

That way, the clone occours only when needed, without forcing the caller
to pre-emptively clone each time it's called.

…binary merges.

Binary merges are mere choices of which side to pick, which works well for big files
as well. Git doesn't define this well during its own merges, so there is some room here.

…merge results.

This works by either selecting a possibly unchanged and not even loaded resource,
instead of always loading it to provide a buffer, in case the user doesn't
actually want a buffer.

Note that this also alters `buffer_by_pick()` to enforce handling of the 'buffer-too-large'
option.

…ration

This also fixes an issue with blob merge computations.

It's breaking because the marker-size was reduced to `u8`.

…tial name.

For instance, previously, a ref named `A` could not be found even though `refs/heads/A` existed.

There it's far more useful and plumbing crates are enabled to write
objects without pulling in `gix-odb` as dependency.

…RT` as far as tests go.

Note that this judgement of quality is based on a limited amount of partially complex
test, but it's likely that in practice there will be deviations of sorts.

Also, given the complexity of the implementation it is definitely under-tested,
but with that it's mostly en par with Git, unfortunatly.

On the bright side, some of the tests are very taxing and I'd hope this
means something for real-world quality.

…writing.

That way it becomes usable when merging trees, which benefits from automatic
checking of hashes before writing loose objects.