Fix locate in let punnings

[panglesd]

This fixes an issue with the locate query reported in #1796.

(Sorry this is quite verbose for such a small PR but it is also here to help me be corrected if I have assumed something wrong)

In a punned let binding, both the pattern and the expression have the same location. None of them is ghost. I think that somehow breaks the assumption that "sibling nodes have disjoint locations".

During a locate query, the first thing is to load the environment at the point of the cursor. The lookup of "nodes at cursor position" will suppose the invariant mentioned above to return a list all nodes the cursor is in (of which we take the first one, the "deepest").

Since two siblings (the pattern and expression) have the same location, the order in which they are considered matter: when the first one is found, we never look at the other siblings, and "enter" it to continue the search.

In the current implementation, patterns are looked first, and we end up with the environment at the pattern to do our "locate". And so, the unhelpful "Already at definition point".

This PR fixes this by folding let operators in a different order, making the expressions side be entered first.

Having the behaviour depend on the order of the fold is slightly brittle, but I think this PR is still fine to solve #1796 without refactoring a huge part to solve a bigger underlying issue ("siblings might have the same location"). But I'm new to the codebase so I might miss something!
(I'm happy to implement a better fix, but then I'd like to discuss with more experienced Merlin disciple first)

[xvw]

Neat!

[voodoos]

I have mixed feeling because this makes the behavior of occurrences (and maybe other queries? we should think about it!) less intuitive. I do think locate is the most important of them so this is an improvement, just one that is not entirely satisfying.

[voodoos]

This is not a fixme anymore, tight ?

Suggested change

	FIXME: Should locate to the `x` in `f x`, not say that we are at the definition
	Should locate to the `x` in `f x`, not say that we are at the definition

[voodoos]

Of course this change impacts other queries too, and some might expect the new value to be considered, not the punned one. I can think of occurrences, but maybe others too ?

Fixing Locate at that costs feels like a fair payment, but ideally we could fix one without breaking the other.

Suggested change

	FIXME Since #2066 this shows the occurrences of the punned value (lines 3 and
	4), not of the pattern (lines 4 and 5). Unlike for the locate query this does
	not feel like the more natural choice. We should think of a more flexible fix at
	some point.
	$ $MERLIN single occurrences -identifier-at 4:7 -filename let_punning.ml < let_punning.ml | jq .value
	[
	{
	"start": {
	"line": 3,
	"col": 6
	},
	"end": {
	"line": 3,
	"col": 7
	},
	"stale": false
	},
	{
	"start": {
	"line": 4,
	"col": 7
	},
	"end": {
	"line": 4,
	"col": 8
	},
	"stale": false
	}
	]