Stats: take into account unused hw axes in run count, refactor code

See merge request !121

Fix PyOpenCLKernelExecutor.get_code to adequately preprocess arg_to_dtype so…

See merge request !125

Fix: Fix PyOpenCLKernelExecutor.get_code to adequately preprocess arg_to_dtype so that get_typed_and_scheduled_kernel caching works

Fix PyOpenCLKernelExecutor.get_code to adequately preprocess arg_to_dtype so that get_typed_and_scheduled_kernel caching works

Minor fixes

See merge request !123

(rather than rely on the PwQPolynomial's domain, which only captures
where the corresponding poly is non-zero)

Merge branch 'stats-with-unused-hw-axes' of ssh://gitlab.tiker.net/inducer/loopy into stats-with-unused-hw-axes

Merge branch 'stats-with-unused-hw-axes' of ssh://gitlab.tiker.net/inducer/loopy into stats-with-unused-hw-axes

Merge branch 'stats-with-unused-hw-axes' of ssh://gitlab.tiker.net/inducer/loopy into stats-with-unused-hw-axes

Preserve TaggedVariable in ExpressionToCExpressionMapper.map_subscripts

See merge request !120

So far, this handler silently dropped any tags on the aggregate,
which is an undesired behaviour if the tag is to be used in the
code generation step.

fixed an issue when trace_assignment_values=True

Cache improvements

See merge request !117

Fix domain tree branch detection

See merge request !115