Add test case for field access vs application precedence

[jfmengels]

I found a few regressions in the pratt-parser branch, so I've written tests for them.

1. Trailing comments at the end of a declaration are a syntax error.
2. f -g a is incorrectly parsed as f - (g a) instead of as f (-g) a
3. foo a.field is incorrectly parsed as (foo a).field instead of as foo (a.field)

cc @jiegillet