Feat/load completion improvements

[scop]

Intended to be merged after #923.

This started from looking into #917 (comment) and ballooned in scope somewhat. The gist of this is that we now provide the full command name (even when not invoked with one) to the completions when sourced, to facilitate generating from the correct executable, and for improved handling of commands invoked with relative paths.

The commits are kind of intertwined, but I hope looking into them individually for details provides enough context if the end result is hard to follow.

[akinomyoga]

Thank you for working on this issue! This is great.

[akinomyoga]

I think we should also fix

diff --git a/completions/_nox b/completions/_nox
index ac387ef7..8e98002a 100644
--- a/completions/_nox
+++ b/completions/_nox
@@ -9,6 +9,8 @@ eval -- "$(
         register-python-argcomplete3 --shell bash "$1" 2>/dev/null
 )"

-complete -p "$1" &>/dev/null
+{
+    complete -p "$1" || complete -p "${1##*/}"
+} &>/dev/null

 # ex: filetype=sh

[scop]

-complete -p "$1" &>/dev/null
+{
+    complete -p "$1" || complete -p "${1##*/}"
+} &>/dev/null

I'm not quite sure about it: I'm not aware of any version of register-python-argcomplete that would register the completion for anything else besides the exact command passed to it. Then again I think we would not even be loading this completion for a full path one if the basename already had a completion installed (because it would have been just run, instead of being looked up), so in that sense the suggested change would not be a problem.

But actually, I thought about it but forgot: this stuff should no longer be necessary here in the first place, because __load_completion now checks if a completion actually got installed anyway. 95eef92 and a44ba64 gets rid of the no longer needed bits.