Failures found using fuzz testing

Using the tools from #379 uncovers several bugs. For example

```julia
fuzz_test(try_parseall_failure, product_token_fuzz(cutdown_tokens, 2))
fuzz_test(try_hook_failure, product_token_fuzz(cutdown_tokens, 2))
fuzz_test(try_parseall_failure, random_token_fuzz(cutdown_tokens, 10, 10_000_000))
```

Some errors found with this:

### Parser errors

* [x] `parseall(Expr, "@(")` (#382)
* [x] `parseall(SyntaxNode, "function(where")` (#388)
* [x] `parseall(SyntaxNode, "x{primitive\ntype")`, `parseall(SyntaxNode, "(primitive\ntype")` (#386)
* [x] `parseall(SyntaxNode, "var\"\"``\$")`, `parseall(SyntaxNode, "+||where'``\$")` (#394)

### Tree building errors

* [ ] `parseall(SyntaxNode, "using . ...", ignore_errors=true)`
* [ ] `parseall(Expr, "\x04'\0", ignore_errors=true)`
* [ ] `parseall(Expr, "@(var\"", ignore_errors=true)`
* [x] `parseall(Expr, "for\n\n<:", ignore_errors=true)` (#393)
* [ ] `parseall(Expr, "x:y:<", ignore_errors=true)`

### Hook errors

* [x] `Meta_parseall("x.")` (#385)

Start with the fuzzing tools from #379 and reproduce the unchecked parser and tree-building cases listed here, including the `parseall` calls using `ignore_errors=true`. Compare them with the checked-off cases and inspect the related issues such as #382, #386, #388, #393, #394, and #385. Done means the remaining failures are resolved and covered by appropriate regression checks.

Written by the indexing model from the issue text.