Fix skipping of SPIR-V test for Nvidia driver

See merge request !137

* make function also work in a subclass

* Use __class__ instead of type()

Co-authored-by: Andreas Klöckner <inform@tiker.net>

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* implement broadcasting array binary ops with device scalars

* test broadcasting array binary ops with device scalars

* re-add some asserts (better to be safe)

* be explicit in error msg

op -> operator

Co-authored-by: Andreas Klöckner <inform@tiker.net>

* array binops: include asserts on out.shape as well

* set default args more elegantly

* Array shape checks: save shapes in temporaries

Co-authored-by: Andreas Klöckner <inform@tiker.net>

Tim Gates authored Jun 30, 2021 and Andreas Klöckner committed Jun 30, 2021

There are small typos in:
- aksetup_helper.py
- doc/types.rst
- examples/black-hole-accretion.py
- pyopencl/__init__.py
- pyopencl/cl/pyopencl-random123/array.h

Fixes:
- Should read `unconditional` rather than `uncondtional`.
- Should read `transferring` rather than `transfering`.
- Should read `recipes` rather than `recipies`.
- Should read `preferred` rather than `prefered`.
- Should read `corresponding` rather than `correpsonding`.
- Should read `arrays` rather than `arrrays`.

Closes #500

* Drop Py2-only Array.__nonzero__

* Adapt docs to absence of Array.__nonzero__

* cl.array.ravel: take order as an argument

* adds test ravel

* Enable flake8-bugbear

* Remove two more attempts to iterate/cast options

https://github.com/inducer/modepy/pull/38#issuecomment-846472970

Extend cl.array.(maximum|minimum|if_positive) to take scalars

- always return 'out' even when the user provides a buffer for it

- probably cleaner to read
- probably better performance (?)

np.number is a Number