cape-ispell just used grep to scan /usr/share/dict/words. We can replicate
`ispell-lookup-words' with very little code thanks to the function
`process-lines-ignore-status'. This way we avoid loading ispell.el. Also the
name `cape-ispell' was rather confusing since it suggested that some kind of
spell-checking is going on, which was never the case.

These utilties can be used to tie a Capf to a certain context. There exist
multiple ways you can organize your `completion-at-point-functions'.

The main programming mode Capf can come first in the list. Then subsequent
auxiliary Capfs will only take over if the main Capf bails out. Sometimes the
main Capf is exclusive and never bails out. You could then turn it non-exclusive
with `cape-capf-nonexclusive'.

Alternatively you could move the auxiliary Capfs before the main Capf. Then you
could use the `cape-capf-inside-*' helpers to make the auxiliary Capfs less
aggressive, such that they trigger only inside comments or strings.

There is no silver bullet here and the setup depends highly on the mode. These
Capf transformers are just meant to give you flexible tools at hand such that
you can build the completion pipeline which works best for you.

(There has been some prior discussion about context dependent Capfs and
predicates a long time ago in #9 and #11, but the addition of this feature got
actually inspired by @LuigiPiucco's Doom PR.)