Fix overflow panic in skipping_iterator_t with explicit saturating_add

[drott]

When shaping extremely long lines with fonts that have classic kern or
AAT kerx tables, the shaper instantiates a single skipping_iterator_t
once outside the loop and reuses it across the entire buffer. Because
self.glyph_data accumulates matching sequence indices without being reset,
it can eventually exceed u16::MAX (65,535) and cause an addition
overflow panic.

This fix upgrades self.glyph_data to a u32, and propagates the type
upgrade to match_func.

Assisted-by: Gemini

[drott]

Not 100% sure if this is the right approach, but does help against issues with encountering overflows with AAT font shaping.

[behdad]

If I understand the code correctly, this should simply be a uint and no saturation needed. In fact, the fact that it's 16bit is one of those beefs I have with this codebase baking a lot of gid16 assumptions around.

[drott]

I'd prefer to avoid machine-sized types like std::os::raw::c_uint and instead have a specific size, but do you mean u64 then or u32 as I did in this change?

[behdad]

u32 is fine, but I prefer to change themay_match and other signatures to pass down u32 as well, all the way. Should need no saturation.

[dfrg]

u32 is fine, but I prefer to change themay_match and other signatures to pass down u32 as well, all the way. Should need no saturation.

+1. Use of unnecessarily narrow types is endemic in this codebase and choosing promotion rather than truncation has been my strategy when I encounter these.

[drott]

Upgraded to u32 and propagating this to match_func PTAL.

[dfrg]

LGTM, thanks!