And reject non-empty paths ending in empty segments.

The following cases were being parsed incorrectly:

• / represents the empty path, ~. This was being parsed into [~. ~]
• /x/ is not valid. This was being parsed into [~.x ~. ~]

This happens because urs:ab has no problem parsing the empty string.
For some supported cases, like //x ([~. ~.x ~]), this is actually desired
behavior, but it results in trailing empty segments for paths ending in /.

Here we apply a +sear on top of the existing parser, that transform the /
case to produce ~, and ensures the absence of a trailing empty segment in
all other cases.

Note that we change (more fas urs:ab) to (most fas urs:ab). Since urs:ab
parses the empty string, this doesn't actually make a difference, but it does
make it more obvious that the +rear call will never crash.

Alternative approaches I attempted all resulted in much more complicated
parser, so the dumb +sear seems preferable.
We do eat the performance cost of an additional list traversal (in +rear)
with this change, but that is probably not the end of the world.

If anyone can golf a cute smol parser that solves for this, I'd be happy to use that instead.

All tests seem to pass, but note that /x/ was previously a valid path, and will no longer parse after this change. Luckily, +stap is used (directly & indirectly) in only a few places:

• parsing for some ford runes,
• +pa:dejs:format (currently only used in hark chat-hook parsing),
• eyre's %subscribe command parsing (which should arguably use +pa:dejs instead).
• scry & export path parsing in pier.c (for -X)

I believe none of these cases expect/care about trailing empty segments, and will continue to work fine with the new behavior.

Fixes #1501.