Correct what the README says about traffic, alerts and speed figures

An accuracy pass over the README, prompted by an external review. Most of it is
wording, but several statements were wrong in ways that mattered.

What leaves the machine:

- The outbound-calls table was described as the complete inventory and was not.
  It now lists the iperf3 server you point it at, the traceroute probes, the
  resolver-identity lookup, the resolver version query, and the alert webhook
  and heartbeat you configure. Two rows understated which addresses are sent.

- "Nothing is ever pushed anywhere" now says "anywhere you did not configure
  yourself". The same claim appeared again three hundred lines later, saying
  data only leaves the box if you scrape it, which is not true of a configured
  digest, alert or heartbeat.

Speed and DNS figures:

- The data-used figure was described as counting every byte a speedtest moved.
  It is what each run recorded, which is a measured floor: framing,
  retransmits, discarded warm-up seconds and the packet-loss probe are not in
  it. A few percent on a normal install, more on one pointed at iperf3.

- The ping figures were described in terms that are true of one engine only.
  Ookla reports a mean over ten samples; iperf3 runs are timed by the daemon
  itself and reported as a median.

- A gap in the DNS line means no lookup in that bucket succeeded. Failures are
  left out of the average rather than dragging it down.

Prometheus:

- Four starter alert rules did not do what their comments said. One fired on a
  single missed sample rather than a sustained outage, one could never fire on
  an installation whose speedtests had never once succeeded, one fired during
  normal quiet periods, and one collapsed the label it was grouping by.

- The metrics list was missing seven series.

Also: rerunning the installer does not change service flags on macOS or
Windows, and now says so; date ranges stop at 366 days from today; the
suggested proxy workaround could not work, because names are resolved before
proxy rules are consulted; the outage heatmap only returns days that have
something to report; and the digest does not bank a reporting period while no
webhook is set.

Co-Authored-By: Claude <noreply@anthropic.com>

Author: cansofgrease