Layout processor should use token IDs, not their String rendering

[andreasabel]

The current layout preprocessor renders tokens as String to check whether they match layout keywords or { } ;. However, each keyword/symbol token comes with an ID that could be checked faster.

Steps to rewrite layout processor (generator) to use IDs instead of String:

• [x] LBNF: check whether { } ; occur in grammar when layout is used.
• [x] Give each symbol a name, like ( is LPAREN.
• [ ] Define token constants in generated lexer for all the keywords and symbols.
• [ ] Reference these constants in the layouter, rather than strings.

[andreasabel]

I chose a different implementation path: Simply use TokSymbol (with the token ID) instead of String in the generated Layout.hs. We need not define constants for each token ID in the lexer. Could be still nice, but this is another issue.