Adjacent broken entries shouldn't be merged into a single Junk

[stasm]

foo = {1foo}
bar = {2bar}

should produce two Junk entries in the AST. Right now they're are merged into a single one because junk is first parsed line-by-line (junk_line) and it's only combined in abstract.mjs. We'll likely need two grammar productions, junk_first_line and junk_follow_line, so that we can distinguish between them in abstract.mjs and only join the first line with its follow lines.

[stasm]

In #211 I implemented an approach which is much closer to how our tooling parsers deal with error recovery: the junk is parsed until a new line which looks like it may be an Entry start is encountered.