When getting tables to truncate, panic on unknown version

[brandur]

This one's related to code review feedback here [1]. When getting a list
of tables that need to be truncated for the TestSchema helpers, it was
possible to receive migration versions that didn't exist and the
function would tolerate it, returning the latest known list of tables.
This often works, but even where it does there's some risk of a new
table coming in that doesn't end up getting cleaned up properly.

Here, panic when receiving an unknown version number so that the problem
can be corrected immediately. The panic may seem undesirable, but these
functions are only used in tests, so it should be fine.

[1] https://github.com/riverqueue/riverpro/pull/281#discussion_r3040295326

[bgentry]

This one's related to code review feedback here [1]. When getting a list
of tables that need to be truncated for the TestSchema helpers, it was
possible to receive migration versions that didn't exist and the
function would tolerate it, returning the latest known list of tables.
This often works, but even where it does there's some risk of a new
table coming in that doesn't end up getting cleaned up properly.

I may have misread the code in question, but I thought it was not returning any pro tables when the version wasn't explicitly listed in the case.