Translation: Resolve conflicts in matching.ml and translcore.ml

[ncik-roberts]

Cherry-picks a commit from #171; once that's merged I'll drop the lambda.ml and lambda.mli changes from here.

Nature of the resolutions:

• (~60% of the conflicts) Upstream added an argument Immediate | Pointer to Pfield in addition to mutable_flag. We added field_read_semantics internally (this is semantically equivalent to mutable_flag, though the constructor names are different). The conflicts are easy to resolve: I just figure out immediate_or_pointer based on the upstream piece of the diff.
• (~3 conflicts) Lazy field accesses are factored out to a function upstream; I add an extra argument to this function because we populated ap_inlined differently at one callsite (though I doubt it actually matters, as I explain in a comment).
• The rest mostly have to do with modes/layouts.

Testing

$ dune b --workspace=duneconf/boot.ws _build/default/.ocamlcommon.objs/native/translcore.cmx
$ dune b --workspace=duneconf/boot.ws _build/default/.ocamlcommon.objs/native/matching.cmx
File "lambda/matching.ml", line 1990, characters 56-71:
Error: Unbound value Obj.forcing_tag
Hint: Did you mean string_tag?

To see that the latter is ok, I recommend adding this module to the top of the file:

module Obj = struct
  include Obj
  let forcing_tag = 0
end

Then it should build.

[goldfirere]

Reviewing

[ncik-roberts]

leaving an unresolved conversation for internal thread

[goldfirere]

#179 was merged into this. Only the first commit needs to be reviewed here.

[Ekdohibs]

Looked at the first commit (matching.ml and translcore.ml) only, looks good.

[ncik-roberts]

Thanks all, merging (not squashing as 1 of the 2 commits in this PR corresponds to a stacked PR).

[ncik-roberts]

(Actually, looks like rebase and merge is disabled and it's not so bad to squash, so I'll do that.)