fix: pwsh autocompletion not returning command hints

[nrynmish]

Closes #266

What's the bug?

PowerShellComplete.get_completion_args() was slicing cwords[1:] to skip the interpreter, but $commandAst.ToString() in the PowerShell template includes both the interpreter (python) and the prog name (main.py) as the first two tokens.

This meant main.py was leaking into the args passed to Click's completion resolver, which couldn't match it to any command and returned no completions — causing PowerShell to silently fall back to path-only completion.

Changes

typer/_completion_classes.py

• Changed cwords[1:] → cwords[2:] and cwords[1:-1] → cwords[2:-1] in PowerShellComplete.get_completion_args() to correctly skip both the interpreter and prog name tokens.

typer/_completion_shared.py

• Replaced $Env:VAR = "" with Remove-Item Env:VAR to fully unset environment variables after completion, preventing stale values from affecting repeated completions.

Testing

Tested by mocking _TYPER_COMPLETE_ARGS and _TYPER_COMPLETE_WORD_TO_COMPLETE env vars and asserting correct args slicing before and after the fix. Also verified end-to-end with pwsh on Linux.

[svlandeg]

Thanks for the PR! I'll put this in draft while the test suite fails, feel free to mark as ready when everything is green.