Fix char-selector parsing when the length is an expression containing a comma

[perazz]

Summary

Char_Selector.match splits the KIND and LEN parts of a char-selector on the top-level comma of the (placeholder-substituted) selector body. The length-first branch restored the placeholder-hidden sub-expressions with repmap() before building the child nodes, but the kind-first and positional branches did not. A length whose value contains a comma — e.g. an intrinsic call — was therefore left as the internal placeholder max(F2PY_EXPR_TUPLE_n), i.e. a MAX with a single argument.

Reproducer

module m
  implicit none
  integer, parameter :: ck = 1
contains
  subroutine s(n)
    integer, intent(in) :: n
    character(kind=ck, len=max(n, 0)), pointer :: tmp
  end subroutine s
end module m

On master this raises FortranSyntaxError: Intrinsic 'MAX' expects at least 2 args but found 1.

The failure only surfaces when every symbol in scope is resolved (all use … , only:). If the module has a wildcard USE, all_symbols_resolved is False, the intrinsic arg-count check defers instead of raising, and the same mis-parse silently leaks the F2PY_EXPR_TUPLE_n placeholder into the parse tree — so character(len=max(n,0), kind=ck) (length-first) always worked while character(kind=ck, len=max(n,0)) did not.

Fix

Call repmap() on the length and kind sub-expressions in the kind-first and positional branches, matching the existing length-first branch. Adds regression cases to test_char_selector.

Fixes #519.