lang: add the readFile built-in

[zerbina]

Summary

Add the built-in readFile procedure to the source language, for
reading characters/bytes from arbitrary files.

Details

• add the semantics for readFile. The content of external files is
  unknown, and thus readFile has an indeterminate result
• add the internal core API readFile and fileSize procedure and
  their implementations as VM host procedures
• implement the source language readFile as a wrapper around the core
  API fileSize and readFile procedures

In order to keep language semantics and implementation easier, at least
for the moment, attempting to read from a file that doesn't exist or
that is inaccessible for other reasons results in readFile returning
an empty sequence, rather than aborting the program.

---

Notes For Reviewers

readFile will most likely not stay as a built-in procedure (whose semantics are part of the language definition), but treating it as one is easier at the moment.

[saem]

Left a comment about making the api a bit more robust, but I'm wondering if that's the right move given that the current incarnation is temporary. If you don't think it's worth it feel free to ignore.

[saem]

would it be too fussy to go down the go route and have it return a tuple (error + seq)?

[zerbina]

I think so, yea. readFile always returning a seq without the ability to reliably detect errors is of course nonsensical in the long run, but having it return a tuple (or union) is tricky.

The first problem is that core.fileSize and core.readFile can fail independently from each other, due to the current inability to lock the accessed file. Reporting an error from the core procedures would require an additional var bool parameter (of sorts) for both, at which point it'd make more sense to use an open/close internal API already.

The second problem, regardless of whether a var bool parameter or open/close API is used, is that both approaches would complicate the implementation of the source language readFile in source2il. The implementation is - in my opinion - quite hard to understand already, and more If guards would only exacerbate this issue. While source2il is going to be thrown out and redone at some point, in my opinion, complexity (and code size) should not grow too much out of hand.

I chose the "return an empty seq when error" solution as a middle ground, where the implementation is simple enough and robust (as in, behaviour is well-defined for all cases), without readFile being completely unusable for the user.