Add allowlist for goals from file nodes

[zrlk]

When goals are read from file nodes, we've seen cases where comments are unintentionally interpreted as containing assertions (even when users provide a goal_regex to avoid this problem). It would be useful to add an allowlist to select which (VName) paths should be searched for goals. (It would also be helpful to clear up the diagnostics printed when reading these goals, as the vname() dump before the bison message looks suspiciously like an internal error.)

[shahms]

Isn't the existing mechanism of reading goals from files specified on the command line, rather than file nodes, essentially this?

[zrlk]

Not exactly. This is the state of the world (and may not be ideal):

Reading goals from file nodes and reading goals from files passed as command-line arguments are mutually exclusive modes of operation. Files passed as command-line arguments can still get vnames from file node entries (done by matching file content), which is necessary to fill in corpus/path/root values.

It's not always the case that a downstream user will easily be able to list the exact files they want to read goals from and/or make those files available to the verifier--consider a test harness that builds an entries file from some target(s) and passes that to the verifier with --use_file_nodes=true. It's simple to say --goal_path_regex='foo/bar/.*', a little less so to enumerate those paths in the harness.

[shahms]

I'm not sure such test harnesses haven't primarily done so out of expedience rather than necessity.