More clearly describe `shares_memory` as being
not always *feasible* rather than *possible*

Also reworded `stride` entry for clarity, moved
'axis' entry to correct alphabetical position.

Also added anchors to reference/arrays.indexing for glossary references
to point to.

Failing CI because it's running doctest on the example strings.

Also, deleting glossary.py, source file of the original glossary.

strctured -> structured, caught by @WarrenWeckesser.

Fixed missing language in `view` entry.

Also adding <BLANKLINE>s to some examples for clarity.

This also allows at least in principle numba dynamically
generated ufuncs to be pickled (with some hacking), see:

dask/distributed#3450

If the name of the ufunc is set to a qualname, using this method,
pickle should be able to unpickle the ufunc correctly.
We may want to allow setting the module and qualname explicitly
on the ufunc object to remove the need for the custom pickler
completely.

This is just a small preparation to make the code slightly more
modular, since I definitely will need to retain the wrapping for
legacy dtype `dtype->f->castfuncs` and the `dtype->f->copyswapn`,
but want more freedom to see whether I can split out the specialized
casts we use internally (since NumPy has those cast functions, but
normally never uses them).

Splitting these two out should make it easier to write a function
to support current "legacy" dtype cast functions without worrying
as much about the casts used internally for NumPy dtypes.

This also includes `NPY_ITEM_IS_POINTER`.  A previous changed
raised an error when this happened, but some downstream libraries
use it to create a custom dtype with a single object field.

It seems acceptable to create such a dtype if (and only if) that
dtype hardcodes names and fields at creation time, so this change
allows that, but otherwise keeps the error intact.

This should work fine, although some care may be required

This allows running `xpress` on current master. `xpress` copies the
type from `np.dtype(object)` rather than using `&PyArrayDescr_Type`.
That seems overall fine, we can just replace it.
The only reason for this check is to ensure that the user does not
override our updated `type(user_dtype)` and thus corrupting it.

This includes a few error paths. Note that the test creates
a new dtype ever time it is run, and dtypes in NumPy are persistent
so that this leaks references.

Also added "contiguous," which is referenced in docs but hadn't been
defined.

Migrated from numpy-tutorials

Migrated from numpy-tutorials

BUG: Fix failures in master related to userdtype registeration

DOC: Use consistent lowercase on docs landing page

BUG: remove `sys` from the type stubs

Reduce the need to scroll, per @mattip suggestion in numpy#17440

DOC: Fewer blank lines in PR template

Remove tutorial wording in favor of article, minor wording updates.

Co-authored-by: Eric Wieser <wieser.eric@gmail.com>

MAINT: Remove duplicate placeholder annotations

BLD: circleCI- merge before build, add -n to sphinx

Co-authored-by: Bas van Beek <43369155+BvB93@users.noreply.github.com>

Co-authored-by: Bas van Beek <43369155+BvB93@users.noreply.github.com>

ENH: Add annotations for bitwise operations

ENH: type np.unicode_ as np.str_

ENH: Added the Final feature for all constants

…delete-install-page

# Conflicts:
#	doc/source/_templates/indexcontent.html