Adding array setting

[henryiii]

This completes the array numpy-like interface, with set support. You can now do:

h = bh.Histogram(
    bh.axis.Regular(10, 0, 1), bh.axis.Regular(10, 0, 1), storage=bh.storage.Mean()
)

# New! Accumulators now support the buffer interface directly!
a = np.full((2, 2), bh.accumulators.Mean())

# New! Set with array
h[3:5, 3:5] = a

If you set with an array, and you leave off one or both endpoints, you can set the flow bins too:

h =  h = bh.Histogram(bh.axis.Regular(10, 0, 1))
h[:] = np.arange(10) # Flow bins are still 0
h[:] = np.arange(11) # ERROR: must match size with or without flow bins
h[:] = np.arange(12) # Flow bins are set
h[0:] = np.arange(11) # Overflow bin is set
h[0:10] = np.arange(12) # ERROR: flow bins cannot be set with specified endpoints

A few notes:

These are not supposed to be deficiencies, just choices.

Dimension adding broadcasting is not supported.

You have to be explicit unless setting with a scalar:

h2[:, :] = 0 # Allowed
h2[:, :] = arr.flatten() # Not allowed, must explicitly add an axis
h2[:, :] = arr.flatten()[:, np.newaxis] # Will broadcast if first dimension matches like numpy

Explicit is better than implicit (and in some cases, Numpy actually does this too, such as when setting a non-complete range on the left).

Mismatched data type dimensions not supported

You can't write

h[...] = 1

If h has a 2, 3, or 4 item dtype (accumulators).

---

Currently, setting a histogram with another histogram is not supported, use h[...] = h2.view().

Removes the old experimental copy_in, this now can be accomplished in Python without the C++ due to the full no-copy views on all storages!

Now only adds the helpful message about compiling if it was truly not able to import _core.

Includes more docs, including a fix for another invalid statement.

Closes #189.