Do not parse past Eof

[cristianoc]

Parsing past Eof can lead to infinite loops, as it violates the assumptions of the termination checker.
See: #540

This PR makes Parser.next assert false when called on Eof. This should not happen as one should check the token before calling Parser.next.

The one exception is during lookahead, for which we provide a nextUnsafe function which does not make progress.

[cristianoc]

This turns potential infinite loops into runtime exceptions when calling Parser.next.
I wad checking the code for exceptions, and realised there are a few parts where there is very nearly an infinite loop (it's an innocent-looking code change away). Which after this change manifests as a runtime exception.
In here #543 the exception analysis is used to make sure this does not happen.

[cristianoc]

Given how noisy it would be to give static guarantees, one can shoot for detection. When look-ahead it taken into account, real code never goes 1 further than the first Eol. This could then reported back as a syntax error message rather than crashing the compiler/editor.

[cristianoc]

Summarising: preventing infinite loops arising from cases where one keeps on reading Eof over and over again statically is a bit tricky right now.
But, it's fairly easy to detect when this happens. The current PR does just that: detection. This will only fire when there's another mistake in the implementation of the parser.
Currently, detection gives an assert false, but perhaps it makes sense to give dedicated error instead. Notice that recovery is not an option in this case (unless one were to use effect handlers in the implementation).

[cristianoc]

At the same time the result of #543 can be used to see where the logic had to be modified, and try to use that to construct infinite loop repros.

Here is one:

let foo = x =>
  switch x {
  | `${

[cristianoc]

Another case to investigate: https://github.com/rescript-lang/syntax/pull/543/files#r897400872

[cristianoc]

Another one: https://github.com/rescript-lang/syntax/pull/543/files#r897402749

[cristianoc]

The last unclear case found during audit does not seem a real infinite loop.
So this should cover all cases considered.

[cristianoc]

Added an issue to track better error messages: #550