Add extra CONV2D local bounds tests

Add specific kernel, dilation and stride tests for checking local
bound (fast path) convolutions.

Fix test selection to understand "type" arg as dtype instead of
"dtype", also fix CTS config usage of "type" in permutes to keep
the same tests selected.
Fix "output_type" usage in CTS config to "out_type", which doesn't
affect tests selected.
Fix test selection to error on unknown arg dictionary.
Fix README about conformance config to remove old options.

Signed-off-by: Jeremy Johnson <jeremy.johnson@arm.com>
Change-Id: Ib077b49f1d3d092267e196fa0d97ef1393816eef