Reset directory on failed file interpretation

[samcowger]

When includeing a script in a different directory, failure to load the file will leave the interpreter's working directory as the directory containing that script. This fixes that by catching any exceptions that are thrown during file load, resetting the dirctory, and rethrowing.

Once #1769 is merged, this behavior should be amenable to automated testing. As it is, the directory is reset on interpreter exit, so neither the failure mode nor its resolution can be encoded in a test script.

[kquick]

The implementation here looks good, I would just make a general observation that in libraries it's not usually advisable to make process-global changes like changing the current working directory because it can have deleterious effects on other threads or expectations of the application using the library. At present this interpreter block probably represents the main thread of saw itself so it's less of an immediate concern, but in general it's probably better to remember target directories (e.g. in a ReaderT context) and then apply them explicitly when doing any file operations.

[samcowger]

This should, for the record, finally fix a bug mentioned in #1341.

[samcowger]

The implementation here looks good, I would just make a general observation that in libraries it's not usually advisable to make process-global changes like changing the current working directory because it can have deleterious effects on other threads or expectations of the application using the library. At present this interpreter block probably represents the main thread of saw itself so it's less of an immediate concern, but in general it's probably better to remember target directories (e.g. in a ReaderT context) and then apply them explicitly when doing any file operations.

Noted and agreed. @kquick and I spoke and agreed that this PR is still appropriate to merge, but that recording this issue would be wise, so I've opened #1791 to surface it more broadly.