Add `append` method to `VaryingDimension` for explicit chunk extension

[maxrjones]

Context:

From review discussion in #3802 (comment), @dcherian suggested adding an append(self, sizes: Iterable[int]) method to VaryingDimension to allow explicitly extending chunk edges.

Current behavior:

VaryingDimension.resize(new_extent) handles growth by appending a single chunk covering the gap:

# [10, 10, 10] resized from 30 to 45 produces [10, 10, 10, 15]

This works but gives the caller no control over how the new region is partitioned.

Proposed behavior:

# Explicit chunk extension
dim = VaryingDimension([10, 10, 10], extent=30)
dim = dim.append([10, 5])  # -> edges=(10, 10, 10, 10, 5), extent=45

# Would raise on FixedDimension (regular grids don't have explicit edges)

Open questions:

• Should this live on VaryingDimension, ChunkGrid, or Array?
• Should FixedDimension.append raise, or convert to VaryingDimension?
• How does this interact with metadata serialization? Appending chunks to a regular grid would change the serialization format from "regular" to "rectilinear".
• Should Array.resize() accept an optional chunks parameter that delegates to this? (Also noted in the design doc's resize section)

Related:

• feat: unified chunk grid with rectilinear chunk/shard support #3802 (unified chunk grid PR)

shall we also add append(self, sizes: Iterable[int]) to allow a different mode of extension (would fail for the FixedDimension).

Originally posted by @dcherian in #3802 (comment)

[d-v-b]

We could define concatenation between these objects:

VaryingDimension([10, 10, 10], extent=30).concat(VaryingDimension([10, 10, 10], extent=30)) -> VaryingDimension([10, 10, 10, 10, 10, 10], extent=60)

But I would only do that if it means we can delete some code.

Should this live on VaryingDimension, ChunkGrid, or Array?

IMO the flow should be:
Array.resize -> new metadata -> new chunk grid -> new dimension objects. So Array.resize needs to handle all the information required to define the new chunk grid.