Hand host setup to hispec install

[mikelangmayr]

• install.sh goes from 235 lines to 71: the root check, the pinned uv bootstrap, uv sync, then exec hispec install. Everything after the virtualenv now lives in instrumentctl and is shared.
• The template unit, polkit rule, bin/hispec and bin/hispec-enable are deleted here and rendered from src/hispec/cli/instrument.toml instead.
• Pins instrumentctl at v0.2.0, where the unit derives each instance's config as /etc/hispec/<name>.yaml rather than reading HISPEC_CONFIG. Every one of the 41 instance files already set exactly that path, so no daemon changes behaviour; the now-unread variable is removed from all of them, and from the docs that described it.
• Adds display_name = "HISPEC", so systemctl status keeps reading HISPEC hardware daemon rather than hispec.
• The hispec-daemon@ migration block is gone: every host is on hispec@. doctor still reports a leftover old unit through the [legacy] config.
• Verified against the pinned version in a clean venv: status -v resolves the config to the derived path, deploy finds and copies it, and doctor reports no stale-variable warning.

Deploy order matters, because the unit and the instance files change together. install.sh renders the new unit before deploying anything, which is the normal flow. The combination to avoid is stripped instance files against a unit that has not been re-rendered, where ExecStart would get an empty -c:

sudo git -C /opt/hispec/app pull
sudo /opt/hispec/app/systemd/install.sh    # renders the new unit first
hispec deploy <names> --no-start --no-enable
hispec restart <names>

[prkrtg]

Good to merge, for future food of thought we might want to think about alternative ways to run the daemons via shell