Expanding "too long" lines should...not

[mattgodbolt]

Some lines are too long to expand. We should avoid doing so.

[ojwb]

mattgodbolt/jsbeeb@8935c9a should have improved the situation here - I think the only problem cases now are where we insert spaces after expanded tokens (which is needed in some cases for re-tokenising to recognise certain tokens - see 37fbf18 where I implemented that).

Maybe a better way to solve the remaining part of this is for "shrink" to remove spaces where we know it is safe to? There's always potential to break programs which peek data within the program, but I think that can already happen with tokenised line numbers vs integer literal line numbers.

I might take a look, unless you think it's a bad idea.