Start reporting errors from the translator

[st0012]

Motivation

First step to #309

I'd like to approach this task with these steps:

1. Since node locations are important for autocorrection, we first make sure parse-trees still match in errored cases. So at this time we don't report errors at the same details & location as Sorbet does, as long as there are errors and they are captured in the fixtures, which we can used for comparison later.
2. Once we got all the parse-trees right, we compare all fixtures under prism_regression/error_recovery/ with testdata/parser/error_recovery/ and try to enrich/calibrate our error reporting.
   • This doesn't include error: Hint: but only real syntax errors
3. Finally, we start working on error: Hint

With that in mind, in this PR:

1. To make it easier to compare with Sorbet's tests, I added error_recovery folder to host error related fixture files as well
2. To make location tests work with 1, which means to accept error_recovery/assign without converting it to error_recovery_assign (can't easily be done in tasks.json), I updated test/prism_location_test.bzl too
3. I added a very basic Translator::reportError function and updated the PM_MISSING_NODE case to start reporting basic errors when parsing invalid Ruby code, which starts with assign.rb

Test plan

See included automated tests.

[egiurleo]

Nice! This is an awesome start. I'm excited to see how it evolves as we handle more errors, and of course curious to see what we have to do to properly handle locations 😬