Refactor shell-hook installation to idempotent ensure semantics (raise on unsupported shell)

[d3mlabs-ai-flow]

From review feedback on #140 (https://github.com/d3mlabs/dev/blob/HEAD/lib/dev/cd/hook_installer.rb):

HookInstaller#ensure_installed currently returns Symbol | false (:added, :already_present, or false for an unsupported shell), which makes for a weak type and models the method's name poorly.

Ensuring the hook is installed is idempotent — a caller of an idempotent ensure has no business knowing whether the work was already done. The contract should be:

• Success = not raising. Drop :already_present; the ensure either leaves the hook installed or raises.
• Unsupported shell raises a typed error (e.g. UnsupportedShellError), caught at a higher level where the user-facing hint is printed.

Scope:

• Dev::ShellRcHook#ensure_snippet (https://github.com/d3mlabs/dev/blob/HEAD/lib/dev/shell_rc_hook.rb) — the source of the tri-state return.
• Dev::Cd::HookInstaller#ensure_installed (https://github.com/d3mlabs/dev/blob/HEAD/lib/dev/cd/hook_installer.rb).
• Callers that branch on the tri-state: Dev::Cd::Accessor (https://github.com/d3mlabs/dev/blob/HEAD/lib/dev/cd/accessor.rb), Dev::Clone::Accessor#announce (https://github.com/d3mlabs/dev/blob/HEAD/lib/dev/clone/accessor.rb), and Dev::Builtins::UpCommand (https://github.com/d3mlabs/dev/blob/HEAD/src/dev/builtins/up_command.rb).

Note: some callers currently print different hints for "just added" vs "present but not active in this shell" — the refactor needs to either fold those messages into one, or surface that distinction some other way that doesn't leak the tri-state into the return type.

Requested by @JPDuchesne in #140 (comment) (thread on hook_installer.rb).