Add regression tests for shell buffer selection ordering

[timvisher-dd]

I thought I'd hit a bug where C-u C-c A to create a new shell followed by navigating away and C-c A would route to the older shell instead of the new one. After investigating (buffer-list) ordering empirically with emacs --batch -Q and tracing the code paths, we proved that the current (buffer-list)-based ordering in agent-shell-buffers works correctly for this scenario: select-window (display-buffer ...) (the path used by agent-shell--display-buffer) does promote the buffer.

These tests document and lock down the actual ordering behavior as regression protection:

• switch-to-buffer and select-window promote a shell to the front
• with-current-buffer (used by agent-shell--handle) does not promote
• bury-buffer demotes a shell to the end
• Buffers created without display stay at the end
• agent-shell-project-buffers filters by CWD and preserves ordering

Checklist

• I agree to communicate (PR description and comments) with the author myself (not AI-generated).
• I've reviewed all code in PR myself and will vouch for its quality.
• I've read and followed the Contributing guidelines.
• I've filed a feature request/discussion for a new feature. N/A since this just tests documented behavior.
• I've added tests where applicable.
• I've updated documentation where necessary.
• I've run M-x checkdoc and M-x byte-compile-file.