Overhaul of vector

[cophus]

What this PR does

This PR completely rewrites the Vector class.

The new implementation is substantially simpler. Public proxy types such as FieldView and CellView are removed, and all indexing/selection operations now return Vector. The explicit NumPy escape hatches are:

• .flatten() for concatenated rowwise data
• .array for selections containing exactly one cell

The main syntax change is field selection. Instead of:
v["kx"]
use:
v.select_fields("kx")

This makes fixed-grid indexing with [] distinct from field selection.

select_fields(...) returns a write-through Vector view, so field-selected updates still modify the parent Vector.

This PR also adds:

• rowwise update helpers such as set_flattened(...)
• append_rows(...) and compact()
• expanded arithmetic support, including **, %, //, unary -, abs(...)
• elementwise NumPy ufunc support such as np.sin(vector) and np.maximum(vector, 0)
• Minor fix for the 2D histogram calculation function.

File I/O is also much cleaner. The internal storage no longer serializes nested per-cell containers, which previously created large Zarr metadata overhead. The new layout stores one numeric row buffer plus per-cell offsets/lengths. In practice this makes Vector saves substantially smaller and reduces read/write overhead.

What PR reviewer(s) should do

• Run the Vector test suite and confirm downstream tests still behave as expected.
• Run the attached notebook examples.
• Call out any missing functionality or awkward syntax before this API is finalized.

Notebook demo for syntax:

vector_new_01.ipynb

[arthurmccray]

Okay this looks good to me. I reckon there will be some changes that need to be made as this gets used in features, but it does everything I want for now at least.

It's not the easiest to understand all of the code, but unless we start subclassing it, then I don't think that should ever be a problem. The tests we have are fairly robust I think.

I also created a tutorial at electronmicroscopy/quantem-tutorials#5