change of the generator to reflect bug fix and improved kernel (from Azzam; Thanks)

add missing restores; fix t30 test - it was assuming allocation initializes values to 0; bug fixes thanks to Azzam; control of templates to be done on CPU or GPU

some defines that control what goes through GPU code

add macro for debugging CUDA kernels

add atomics.cuh in the backened. Before was in magma and this was creating problem for spack

backends: add missing CeedVectorRestoreArray calls

Merge branch 'master' of github:dmed256/libceed [PR #62]

* 'master' of github:dmed256/libceed:
[occa] Updated verbose flag

occa: fix memory leaks

Valgrind still says there is memory lost under some of these (plus a lot
from Occa static constructors), but this at least makes an attempt to
free from our end.

[occa] Updated verbose flag

Merge branch 'master' into magma-dev

occa: fix memory corruption from unterminated string

TODO: Need to define CeedDebug...

Merge branch 'master' into magma-dev

Conflicts:
Makefile
tests/tap.sh

occa: use CEED_INTERN for internal backend functions

[occa] ceed-occa-okl file, CeedOklPath for qfunction too

[occa] OKL file occa_cache_dir & libceed_dir check

[occa] OKL file search refactoring, error checks & dladdr no ifdefs

[occa] style, typo & comments

[occa] use dladdr to find restrict, basis OKL files and look in the OCCA cache for user's qfunctions

[magma-dev] Tweaks in Makefile. Fix formatting to suppress warnings.

[occa] PETSc example with ctx inout, qfdatasize & offsets

[occa] device from ressource, OpenCL TILE size tied to 1

Copyright banners, few other small changes

make style

[occa] fixed /ocl/occa