[FEA] Investigate if release_assert impacts performance

[jrhemstad]

Is your feature request related to a problem? Please describe.
release_assert is used throughout libcudf code to signal an error in device code.

Most concerning is that this is used in type_dispatcher error path.

It's come to my attention that using the __assertfail may incur syscalls, which can force ABI constraints that result in a performance impact even in the happy path.

Describe the solution you'd like

We should benchmark with and without the release_assert in the type dispatcher (using the type dispatcher benchmark to start) and see if there is any difference in performance.

[jrhemstad]

#7583

Turns out, it does! Kind of. It has a pretty sizable impact on binary size and compile time. Actual runtime performance impact seemed more minimal.

[harrism]

Should we close this now?

[jrhemstad]

Yeah, I think it's fine to close this now given that #7583 made it so the assert is no longer present in Release code.