Consolidate concat_dim earlier in prepare_target

[TomAugspurger]

Closes #227
(maybe, might be worth testing).

A few notes:

1. I'm not really happy with how I'm handling the locking (see https://github.com/pangeo-forge/pangeo-forge-recipes/pull/229/files#diff-e12c886cc124886c5cfa5313d760a36c39649af9da845077c663e6feab8487b5R245-R253). I just kind of overwrite what the "real" lock determination code does. This feels too special-casey, but I haven't dug into the how that conflict code works.
2. I haven't tested this at scale yet. Maybe @sharkinsspatial can help verify that this actually fixes things? Let me know if you're able to, or if I should get something set up.

[TomAugspurger]

The diff here looks strange.

All I did was

1. move the if config.consolidate_dimension_coordinates to its own function so that I could call it from two places (here and earlier in prepare_target for concat_dim)
2. Let the caller pass in the dims to check (so that we can do just concat_dim early on, and the other dims later; I think this is what we want).
3. Added a check that arr.chunks != arr.shape (to avoid rewriting concat_diminfinalize_target`, since it would already be consolidated)

[sharkinsspatial]

@TomAugspurger I've been locked up on other things today but I'll try to run some testing with this tomorrow 👍

[rabernat]

/run-test-tutorials

[TomAugspurger]

As discovered by @sharkinsspatial, this implementation buggy (sorry). When we call resize, we end up with a "fragmented" array that has 360,000 entries and a chunksize of 1. So it's already too late. I'll explore our options (and the CI failure)

[rabernat]

@TomAugspurger - over in #236, I have removed the pre-initialization of the concat dim with 0 values. It doesn't seem to break anything (including the tutorial notebooks). If that is the main bottleneck, I would prefer that (simpler) solution over this (more complex) one.

I think the real fix for #227 is upstream in fsspec, related by supporting batch_size more broadly across async operations.

[rabernat]

p.s. If you plan to keep working on this branch, please rebase as there have been some big improvements to the test suite.

[cisaacstern]

Closing as stale/out-of-date. Thanks all!