Text format: memarg doesn't follow lexical format

[tlively]

The offset and align fields in memargs are specified as 'offset='o:u32 and 'align='a:u32 with no space between the keywords and the u32s. However, according to the longest match rule in the lexical format, valid offsets and alignments such as offset=0 and align=3 are reserved and should not appear in the source.

Obviously we can't fix this by changing the format of the offset and align fields due to backward compatibility. Is there a way we can safely fix this in the lexical format?

[rossberg]

Yeah, the spec hand-waves about this by saying

Lexically, an offset or align phrase is considered a single keyword token

in the section on memory instructions. Of course, that contradicts the earlier claim that all keyword tokens are defined by literals in the grammar. I suppose the simplest fix is to widen that statement appropriately.

[tlively]

Oh good, I hadn't seen that text. I'll make a PR adding more nuance to the definition of keyword tokens.