Implement [[ fragment literals ]]

[Rangi42]

Fixes #500

Replaces #716 and #1603

Anywhere you would normally use an address, such as dw Label or ld hl, Label or call Label, you can use a [[ fragment literal ]] instead. Each fragment literal becomes its own SECTION FRAGMENT to be appended to the current section.

[Rangi42]

Summarizing some points from the discussion in #500:

• You can write single-line fragment literals like [[ db "Hello world!" ]], or put the contents on their own line(s), following whatever indentation style you prefer.
• Since these literals are implemented as section fragments, they get appended after their containing section, but any exact placement beyond "somewhere in the same bank as their containing section" is not guaranteed and should not be relied on. (For example, using jr across fragment literals is a bad idea.)
• The parser requires end-of-line tokens as line terminators, not separators, including directives which must manage their own newlines and are "DEFINITELY one of the more FRAGILE parts of the codebase" (so refactoring them to be separators would be a lot of hard work). This means that lexing the ]] token has to inject an end-of-line token before itself, to allow ]] on the same line as its content.
• ASMotor has had these since 2019 as "code/data literals". The implementation is similar to this PR, creating a new section for the literal content, placed in the same "group" as its containing section, and using a special label to evaluate the literal content's address.

[Rangi42]

Since this is a backwards-compatible feature, I think it's fine to add in a patch-level release (i.e. 0.9.1 instead of 0.10.0). We've done so before, like with 0.6.1 adding rgbgfx -o - - support, 0.5.2 adding compound assignment operators like DEF x += 1 (and also deprecating SET in DEF x SET x + 1), 0.5.1 adding CHARLEN and CHARSUB, etc. It's just that more recently, we've had a slower release cadence and "minor" releases (more like major ones, it's just that we're still stuck in 0ver) have felt more appropriate.

[Rangi42]

Note that one of the initial blockers to this feature was the global parser state that fragment literals had to back up. We've long since gotten rid of that, consolidating them into the stacks of lexer state and fstack nodes.

[Rangi42]

If anyone wants to build this and test it out before it gets merged -- notice any bugs, inconvenient limits, nice use cases, etc -- that would be very helpful!