Increase dynamic precedence of call

[nmote]

This is a weird one.

I found that under certain circumstances, the pattern foo(1, 2) was not matching the source file foo(1, 2) in Semgrep. When I looked at the CST generated by tree sitter, I found that if foo(1, 2) is followed by a newline, it is parsed as a call, but if it is not, it's parsed as a constructor. Often, patterns are provided over the command line and do not include a newline, whereas source files usually do. I tried to reproduce the difference in parsing using tree sitter directly, and could not. This puzzled me. I considered the possibility that the extensions we've made to the grammar to parse Semgrep constructs might have affected this, but that seemed far-fetched to me.

I recalled that calls and constructor calls are truly ambiguous with each other, so I decided to start looking into how they are resolved. There is an entry in the conflicts array for _simple_user_type vs _expression, which is where the conflict ultimately manifests. However, there weren't any relevant calls to prec.dynamic which would give the parser information on how to resolve the conflict at runtime. So, I formed the theory that tree sitter arbitrarily resolves it in one direction, whereas ocaml-tree-sitter's generated code sometimes arbitrarily resolves it in the other direction. I made the change here, propagated it over to Semgrep, and now have the desired behavior where Semgrep parses foo(1, 2) as a call whether or not it is followed by a newline.

Test plan: Automated tests, plus manual testing with Semgrep described above.

[alex-pinkus]

Great deep dive, thank you for doing this!