docs/roadmap.md says CSV is 'one file per day', --help says 'one per run'

[evandhoffman]

docs/roadmap.md and monitord --help describe the CSV rotation differently.

docs/roadmap.md:

writes rotating, human-readable CSV — one file per day, hostname in the
filename and as a column

monitord --help (v1.6.2):

Files are named sensors.<host>.<date>_<time>.csv, one per run, so files
from …

Both are true at once — a file opens when the daemon starts, and again at local
midnight — but read side by side they contradict, and the filename format
(<date>_<time>) matches the help text rather than the roadmap.

A reader deciding how to consume these files needs to know both rules, because
they determine different things: "one per run" says a restart splits the data,
"per day" says a long run splits it at midnight. Someone who reads only the
roadmap will not expect a restart to start a new file, and someone who reads
only --help will not expect a file to end at midnight.

Suggested wording, for whichever place is canonical:

a new file per run and at each local midnight, named
sensors.<host>.<date>_<time>.csv

Minor, and last of the four filed tonight.

Observed while consuming a 13-hour capture: one run produced exactly one file
(sensors.macbook-pro_local.2026_09_05_08_51_36.csv, 15.5 MB, 08:51-22:06) and
did not roll at midnight, because the run did not cross one — consistent
with both statements, which is why the ambiguity is easy to miss in practice.