help.lint.coverage: fix coverage on predicate? words
fix a regression for `predicate?` words revealed by uses of
`"help" word-prop` over `help:word-help`.
`"help" word-prop` is `f` for `predicate?` words like `array?` and
`zero-matrix?`. a poorly-written comparison in
`M\ word <word-help-coverage>` meant that result was effectively
compared to an empty array { }, instead of just being tested for
emptiness.
additionally, `(word-help)` is introduced to revive the old behaviour of
`help:word-help` before the move to split `$inputs` and `$outputs` in
Help rendering. `(word-help)` provides the actual help content for
words where `"help" word-prop` is `f`, by calling `word-help*`.