current-exe-mismatch.rs test may be too eager to ignore its failings

[Felix S Klock II]

Spawned off of #488 (comment)

The newly added current-exe-mismatch.rs is a bit over-eager to categorize problems as "ignorable" rather than providing some way to flag them as a hard-error in certain contexts.

The main example: while it is not reasonable to expect readelf to exist on every target, it may be reasonable to assume its available on Linux (yes?), and thus maybe a readelf invocation failure should be ignored on non-Linux systems, but a test failure on Linux systems. (Or maybe just treat it as a test failure on Linux Continuous Integration hosts, which is what we really care about here...)