Fix some bugs in ledger-complete-at-point date completion

[bcc32]

ledger-complete-at-point currently seems completely broken for date completion
(e.g., "19" completes to the 19th of the current month, or the previous month,
whichever date is in the past from today).

I added some regression tests that demonstrate the bugs, and then separate
commits to fix each bug. I plan to squash the test and fix commits together
when merging.

(@purcell, I spent some time fighting with the CI over some old calling
conventions for functions (func-arity, which didn't exist; and encode-time,
which doesn't support a single argument in Emacs < 27, but the multiple-argument
form is marked "obsolete" in Emacs 29). What's ledger-mode's policy on
supporting Emacs major versions, is there generally some guideline we follow as
to when we're okay with abandoning support for older versions?)

Fix #419.

[purcell]

What's ledger-mode's policy on
supporting Emacs major versions, is there generally some guideline we follow as
to when we're okay with abandoning support for older versions?

I don't think it would be controversial to only support the latest three major versions, ie. dropping support for Emacs <= 26. I don't know if anyone is likely to declare a policy around this though.