Rewrite most of the shell detection stuff (do not merge)

[uranusjr]

The goal is to fix #915 (comment)

[uranusjr]

Manually tested on cmd.exe, Powershell (pwsh.exe, but powershell.exe should work), and Bash in WSL. This needs more testers.

[uranusjr]

Some explaination.

I did some process-digging and it turns out there is not a trace of Cmder.exe anywhere in the process tree when _win_utils.get_shell() is called, so I removed it from SHELL_NAMES. It is not useful anyway, since we really need to know not only we’re running Cmder, but also what shell it is spawning for us.

There are quite a few variants of subshell-spawning, each requiring a different call:

• Non-bash shells on non-Windows (simplest to handle).
• Bash (not including Msys Bash, which never worked).
• Bare cmd or Powershell (including pwsh)
• cmd in Cmder
• Powershell in Cmder

The logic gets complicated pretty quickly, so I decided to refactor them into five classes (plus an intermediate for the Cmder cases). _detect_shell now returns an instance of Shell (or its subclass) for shell to use.

[techalchemy]

The decision tree in this function is kinda messy, wonder if we can clean this up any

[uranusjr]

Rewrote this a bit. Not sure if this makes more sense.

[techalchemy]

Overall these changes make sense. I think we are maintaining 2 lists of windows compatible shells, we should consolidate those. Otherwise I think we can schedule this for 11.11

[uranusjr]

TODO:

• Make sure this does not overwrite Escpace spaces on cmder shell path - Fixes #2115 #2168.
• Generate patch.

[hstefan]

Comment on #915