The test data needed for this, along with some yak-shaving notes, are
on the 'kff-debugging-sent-recv-count-weirdness-2024-04-02' branch.
That branch should probably be resolved and merged before this
functionality is finished.  See also private:r12349.

The diff looks big but is's mostly an indentation change.  The
semantic change is just wrapping a `max' call around the calculation
of the number of days since this recipient was sent to.  That way if
the number of days is zero it gets bumped up to a tiny number, and we
thus avoid a latent divide-by-zero error (that bug is currently masked
by a sqrt call, but we shouldn't rely on that accidental protection of
course).

We should distinguish between sent-to dates and received-from dates,
for better scoring.  This commit just adds a comment that saves my
progress so far in figuring out where and how to do that.

Also, remove material from the README (added in commit ad8ed228)
about needing a better ranking algorithm.  Since commit 2d9f2133
we have that; date-distinguishment will just improve it.

Also add new user option `mailaprop-show-scores' to control whether or
not to display each candidate's score next to the sent/received counts.

Apparently I made this change on March 8th and forgot to commit it.

Response time was too slow when we offered completion starting with
any input whatsoever.  Requiring a minimum amount of initial input
before company-mode starts offering candidates speeds things up.

Have `mailaprop-load-addresses' prompt for a filename when invoked
interactively with a prefix arg.

Fix reversed_unquoted_name_re so that when parsing a full address like
"Lastname, Firstname Middlename <flastname@example.com>" we don't drop
"Middlename".

While this new test data is ready, the bug that it reproduces is not
yet fixed.  Therefore the test data has a ".wip" suffix, so it won't
be included in the default './check' run.  We'll remove that suffix
when we fix the bug.  To reproduce the bug, do

  $ ./mailaprop.py --suppress-header              \
    < test-data/bonkers-backslashes-etc.mbox.wip  \
    > foo.out

followed by `(mailaprop-reload-addresses "foo.out")' in Emacs, and you
should see an error that looks something like this:

  +: Wrong type argument: number-or-marker-p, " \
  <someone@michaellacroix.com> <levinson@domaintwo.edu>"

This was always my intent; I'd just typo'd the `case-fold-search' binding.

Also make a minor clarification later about interactive speed.

This is easier for me to remember and avoids a completion collision.

We now handle the following exceptions when parsing headers:

  * "UnicodeEncodeError: 'utf-8' codec can't encode character '\udcf8'
    in position 64: surrogates not allowed (also from absorb_headers()"

  * "email.errors.HeaderParseError: expected ':' at end of group
    display name but found '@GMAIL.COM"

    (which would then lead to "TypeError: 'ValueTerminal' object does
    not support item assignment")

  * "ValueError: hour must be in 0..23" (from bogus Date headers)

Also:

  * Solve a newly-discovered quoting problem in the output

  * Ingest 'reply-to' as an address-containing header.

This is part of issue #1.

This is part of issue #1.

This follows up to commit ce519dab.