Show how much moved, not only how fast

[evandhoffman]

Closes #35.

A second number in every legend entry: how much moved over the window the
chart is drawing, beside how fast it is moving now. Network and disk in
bytes, packets and disk operations as counts.

┌─ Network ─── 2 min ──────────────────┐
│  ● In     245 Mbit/s        1.4 GB   │
│  ● Out     12.4 Mbit/s       88 MB   │
│ ▁▂▅█▆▃▂▁▁▂▄▇█▅▃▂▁▁▂▃▅▆█▄▂▁▁▂▃▄▅▃▂▁   │
└──────────────────────────────────────┘

The sum is already in the buffer

A rate sample is the mean over the gap before it — RateTracker produces
Δcounter / Δt — so Σ(rate × preceding gap) telescopes back to exactly the
counter delta. WindowTotal is that sum. No counter history is kept, no
source changes, and nothing is read that the ring buffer does not already
hold, so this stays well inside the rule that nothing is written to disk.

The round-trip test is the whole premise: build a synthetic counter,
differentiate it with RateTracker, integrate it back, expect the original
delta. Deliberately uneven intervals, so it cannot pass by multiplying by a
constant it guessed right.

Three things it refuses to fake

• It reports the span it covered. Ten seconds after launch the buffer
  holds ten seconds, and "2 min" over a twelfth of that looks exactly like the
  right answer — which is why nobody would check it. The span sits beside the
  title and reads the window only when the card really has it.
• A gap nobody sampled is not traffic. An interval wider than
  maximumGap is clipped and the remainder comes off covered, so a laptop
  back from sleep cannot credit one sample with an hour that never crossed the
  wire. AppModel.totalGap is four ticks of the master clock, so it follows
  the Sampling tab rather than assuming 0.5 s.
• Fewer than two samples has no total. Nil, not zero — the same answer
  RateTracker gives on a first read, for the same reason.

Network totals in bytes

The rate stays Mbit/s because a link is quoted in bits. A volume is quoted in
bytes everywhere it matters. MetricUnit.accumulation holds the ÷8 as its
only definition, the way NetworkSource holds the ×8 — derived from the unit
like direction and composition, so a source added later gets a total
without anything being told it exists.

The one thing worth arguing about

showsTotals is off by default, following mirrorsPairs and
stacksParts. The precedent's reason is that those change what the picture
means; this only adds a number. The narrower reason it still applies: it
adds one to a header already decided by ViewThatFits, so switching it on
reflows cards onto two lines — a card changing shape on upgrade. Happy to
ship it on instead and say so in the release notes.

Non-goals, each deliberate

No totals row in the CSV (a spreadsheet's SUM × interval is this same
arithmetic, and a "Total" row under the Time column is a type error waiting to
be pasted into a chart). No odometer tile — considered, and it is very much
this app's idiom, but it costs panel space and a third tile kind in
PanelArrangement. No monitorctl totals; no watt-hours; nothing past ten
minutes until the store lands.

Verification

swift build && swift test (205 tests), swift build -c release, and
swiftformat --lint all clean. Not yet looked at in swift run monitor — the
open question is whether a legend entry carrying two numbers still reads at a
glance, and no test can answer that.

https://claude.ai/code/session_01EqTTtmt4fyNtMxjc2ZVBzj

[evandhoffman]

Folded into #38, which carries these commits unchanged. Splitting this off bought nothing: the stack was only ever mergeable in order, so it was three squash-merges and three releases for one change.