Reject reversed position ranges in parser/lexer entrypoints

[ksss]

Summary

alloc_parser_from_buffer and alloc_lexer_from_buffer already rejected negative positions but accepted ranges where start_pos > end_pos. Such a range made the lexer's main loop condition (current.byte_pos == end_pos) unsatisfiable: current advances past end_pos and the equality is never reached, producing an infinite loop in the C extension with the GVL held (SIGINT ineffective).

Reproducer

RBS::Parser.parse_type("", byte_range: 1..0)
# hangs forever; only `kill -9` stops the process

The same hang is reachable from RBS::Parser.parse_method_type via its byte_range: keyword, and from the low-level _parse_signature / _parse_type / _parse_method_type directly with reversed start_pos / end_pos arguments. The _lex entrypoint always passes start_pos = 0, so the new guard on the lexer side is defensive rather than a present-day bug fix.

Fix

Add a reversed-range guard alongside the existing negative-position check, and extract both into a validate_position_range helper shared by alloc_lexer_from_buffer and alloc_parser_from_buffer:

static void validate_position_range(int start_pos, int end_pos) {
    if (start_pos < 0 || end_pos < 0) {
        rb_raise(rb_eArgError, "negative position range: %d...%d", start_pos, end_pos);
    }
    if (start_pos > end_pos) {
        rb_raise(rb_eArgError, "invalid position range: %d...%d", start_pos, end_pos);
    }
}

Reversed ranges now raise ArgumentError immediately instead of reaching the lexer.

---

This PR description was written by Claude Code.