Add server/resetCache and use it in the server tests

[Simn]

I would like to do this because

1. it should be faster and
2. any problems that pop up here suggest that we're not resetting some state.

There are some failing test and I had to comment some of them out too because they raised exceptions. I don't know how people debug node.js applications without becoming alcoholics.

[Simn]

Observation: The failing tests are all in ServerTests:

assertations: 808
successes: 799
errors: 0
failures: 9
warnings: 0
execution time: 36.411

results: SOME TESTS FAILURES (success: false)
cases.ServerTests
  testDiagnosticsRecache2: FAILURE .F
    line: 251, expected true
  testDiagnosticsRecache3: FAILURE ..F
    line: 265, expected true
  testSyntaxCache2: FAILURE F
    line: 307, No such completion
  testSyntaxCache: FAILURE FFF.
    line: 276, No such completion
    line: 280, No such completion
    line: 289, No such completion
  testXRedefinedFromX_2: FAILURE FFF.
    line: 479, expected true
    line: 481, expected true
    line: 483, expected true

But running with -D UTEST_PATTERN=ServerTests:

assertations: 112
successes: 112
errors: 0
failures: 0
warnings: 0
execution time: 9.794

results: ALL TESTS OK (success: true)

This means that some other test messes up some shared state.

[Simn]

I just noticed let fake_modules = Hashtbl.create 0 in typecore.ml which makes me realize that I didn't check for any global Hashtbl things.

[Simn]

Another one is Lexer.all_files, but I think that one is actually supposed to be global information because it resolves position offsets to line/column information.

... which actually makes me wonder how this is supposed to work with hxb where we don't always have any lexer information to use.

[Simn]

Merging hxb didn't change much (which is good). OCaml 5.0.0 still fails, the rest works fine now.

@kLabz I'm assigning to you because I don't know what else to do.

[Simn]

So the problem here is the test for #10646. For some reason, the ContextMemory request fails on OCaml 5. That's not a problem with this PR specifically and since the changes here should be good, I'll merge and reopen #10646 to make sure we add back the test in a way that is compatible with both OCaml 4 and 5.

[Simn]

So speed-wise, nothing much changed after all. I thought that I tested this before and got a noticeable improvement, but that doesn't appear to be the case.