Cmap format 4 subtable length check is too strict

[mikeday]

The cmap format 4 subtable ends with an array of unspecified length, and Allsorts determines its length by checking the length field for the subtable given earlier. This should always be even given that the subtable only contains u16 values, however we have encountered fonts in the wild for which it is odd -- Allsorts will refuse to process them while other implementations will accept them.

I think we need to just round down the length here, and potentially log a warning if warnings are enabled, rather than refusing to process the subtable. (Any other similar length checks may also require scrutiny).

[mikeday]

Font encountered here:

terrastruct/d2#1252