Fix pytest completion list separators

[IanHollow]

Summary

Two pytest completion providers use semicolons inside list literals. On Nu 0.115.1 this silently returns only the first value; on Nu 0.116.1 importing the completion module fails with Unexpected semicolon in list. Replace those separators with commas, matching the neighboring providers, and add a standalone regression that sources the real completion file and checks the intended values.

-["w"; "a"]             ["plain"; "rewrite"]
+["w", "a"]             ["plain", "rewrite"]

Evidence

On native x86_64 Linux, using Nu 0.115.1 and 0.116.1:

• Before: the same standalone regression fails on both versions. On 0.115.1 the log-mode provider returns [w] instead of [w, a] and completion omits a; assertion completion similarly omits rewrite. On 0.116.1 importing the original file fails at the semicolon.
• After: module parsing and the standalone regression pass. Both provider lists and actual commandline complete --detailed results include all intended values.
• nu --no-config-file --commands 'use toolkit.nu; toolkit check pr --and-exit' passes on both versions.
• nu --no-config-file custom-completions/pytest/tests/pytest-completions.nu passes on both versions.

The regression script is run explicitly: toolkit test is currently a stub, so a successful toolkit check pr alone does not prove the assertions executed.

Assistance: OpenAI Codex implemented the change and regression, ran the recorded checks, and drafted this description.

Merge Danger

Door: two-way. Blast Radius: pytest completions. No pytest invocation or option semantics change; the providers return the same intended values.

[fdncred]

Thanks