* fix(costs): keep the daily average for a measured-zero range

averageDailyValue returned null whenever the accumulated total was zero,
so a cost history window the CLI measured as zero throughout lost its
"Average/day" line entirely. The history rows built from the same points
print those zeros since #274, and the function already counts a measured
zero day in its denominator, so the summary contradicted the rows right
below it. A provider with no priced usage in the window, such as a local
model, hit this on every visit.

Gate the null on an empty denominator instead of a zero total. That keeps
a range with no measured day without an average and avoids the 0/0 the
bare total check was hiding.

Constraint: peak selection stays metric-based and still names no day when nothing was spent
Rejected: returning zero from peakPoint too | its comment and test pin the empty peak deliberately
Confidence: high
Scope-risk: narrow

Documentation impact: CHANGELOG Unreleased gained an entry; docs/usage.md
and docs/cost-history.md describe the measured-zero average. TODO is
unaffected: presentation-only change, no CLI contract touched. No new
user-facing string, so the catalogs are unchanged.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* test(costs): pin the formatted zero the average line prints

The average test asserted the numeric result only, so a regression in the
formatters could still print "Average/day: -" for a range measured as
zero. amountString returns the dash for anything that is not a finite
number, and no test covered either formatter at zero.

Assert the two figures the caller composes the line from.

Confidence: high
Scope-risk: narrow

Documentation impact: none; test-only change with no behavior difference.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>