record punning & record patterns

[leque]

closes #358 and #359.

This PR implements record punning and record patterns in 0.1.0 branch.

Record punning is a short-hand notation for record construction. You can omit a field value if it is a variable reference and the variable has the same name as a record field. E.g. (| x = x |) can be written as (| x |).

Record pattern enables us to pattern match against record values. Record patterns are typed as described in #358:

  % fx : (| x : 'a |) -> 'a
  val fx (| x |) = x

  % fx- : (| x : 'a | ?'b |) -> 'a
  val fx- (| x, _ |) = x

Some remarks:

• --bytecomp routine is implemented, but is not tested since 0.1.0 does not support eval_libary_file in bytecomp mode yet.
• Exhaustiveness check is not implemented because I do not understand what algorithm is used in Exhckecker. It would be nice to implement it as a separated PR.
• This PR turns tests/parsing tests into inline tests to prevent the possibility that inserting tests changes locations of other tests.

[gfngfn]

The description COMMA? allows (| foo = x, _, |) for a record pattern as well as (| foo = x, _ |). Is this intended?

[leque]

Yes, this is intentional. I think there is no reason to forbid a trailing comma after a wildcard.

[leque]

FYI: OCaml also allows ; after _ in record patterns, but forbids ; after .. in object types; i.e. {x; _;} is acceptable, <x: int; ..;> is not. This PR's syntax is derived from OCaml's record pattern syntax.

[Et7f3]

Same for list [ value1; ] python also accept trailing comma: one advantage is we can swap item and it will just works. Other reason seems to be human better deal with terminator than separator (last argument was given by an ocp-indent author).