Say why the dashboard refused a run, and accept "never" and "1 hour"

[ntatschner]

Say why the dashboard refused a run, and accept "never" and "1 hour"

Starting a run from the dashboard said "That is not a run this can start

• a template cannot be run, only copied." for teams that were not
  templates. The page maps every exit code 2 from team run to that one
  guess, because only the exit code crosses back from the command; the
  words go to the terminal serving the page. The real refusal was the
  new "Take the lead's recommendation after" box: its placeholder reads
  "never", so people type never - or 10, or 1 hour - and none of those
  parsed as a duration.

Reproduced against 0.42.0 through a sandboxed daemon, with no agent on
PATH so nothing could launch: "never", "10" and "1 hour" in that box
each gave the template message; an empty box got as far as the agent
check.

• never, off, none and no mean no timed answer. For a run they
  override a team's own setting, and the page passes nothing for them.
• Durations may spell the unit out: 30 mins, 1 hour, 2 days, 1.5h. A
  bare number is still refused, since ten of one unit is not ten of
  another.
• An unreadable duration is refused by the page before it starts
  anything, in the same words team run uses. The fallback for any
  other exit 2 no longer mentions templates, which the page never
  offers, and points at the terminal instead.
• A criterion or goal typed as a list item ("- tests pass", "1. ...")
  broke the whole run: a value starting with a dash is read as an
  option ("Option does not have a name"). The marker is taken off, and
  criteria travel as --done-when=... so one that truly starts with a
  dash still arrives as a value.

StartFormTests gains tests for each, including one that parses the
page's own argument list with team run's settings class and checks the
goal, the criteria (one starting with a dash) and the duration arrive
intact. Two existing assertions changed from the separate
"--done-when", value pair to the joined form; their point - one option
per criterion - is unchanged. Put back to main's DashboardActions, the
new and updated tests fail (7 of 23).