Disabled runs and silently-broken runs print the same summary

[pftg]

Visual regression is expensive, so teams run it sometimes — like SimpleCov. The gem supports that (SnapDiff.config.enabled = false short-circuits assert_matches_screenshot, and #279 put the readiness block inside the same guard so it costs nothing either). What it does not support is knowing which mode you are in.

The defect

A run with visual checks deliberately OFF prints:

[snap_diff] 0 verified, 0 changed, 0 new (not verified). NOTHING WAS VERIFIED -- no screenshot was compared to a committed baseline.

A run with visual checks ON that compared nothing — the exact bug 2.0 exists to kill (#255, #256) — prints the same line.

Reproduced:

SnapDiff.config.enabled = false
SnapDiff::Reporting.send(:counts_summary)
# => "[snap_diff] 0 verified, 0 changed, 0 new (not verified). NOTHING WAS VERIFIED -- ..."

Two opposite situations, one message. Worse: developers who run with screenshots off see this constantly and learn to ignore it — which trains them to ignore it on the day it means something.

Proposal

1. When active? is false, say so instead: [snap_diff] visual checks OFF -- 14 assertions skipped. Enable with SNAP_DIFF=1.
2. Count skipped assertions, so the size of what was not run is visible.

Both are cheap; the counter is one increment in the existing early return.

Why this matters more than it looks

Real usage from a consumer that runs the suite this way, jetthoughts.github.io: tiers are hand-maintained regexes of test names —

CRITICAL_TESTS = "/homepage|blog_index|blog_post$|visit_blog_post|course_|..."
SMOKE_TESTS    = "/test_homepage$|test_blog_post$|test_not_found$|..."

That list is a maintenance surface with the same failure mode: rename a test and it silently drops out of the gate, with nothing reporting the loss. Their comments also record one test at ~196s being 44% of the suite and excluded by hand.

Making "what did I skip" visible is the smallest change that helps; it does not require the gem to know anything about test selection.

Out of scope

Selecting which screenshots to run from a diff. The gem cannot know which code affects which page, and guessing would be worse than the current explicit lists.

[pftg]

Researched against SimpleCov (#276) — this is the first of three steps, and the smallest.

Scope here: when active? is false, say so and count what was skipped.

[snap_diff] visual checks OFF -- 14 assertions skipped.

instead of the current line, which is byte-identical to a suite that should have been comparing and wasn't.

The counter is one increment in the existing early return in assert_matches_screenshot; the message is one branch in counts_summary.

Deliberately not here:

• SNAP_DIFF=0/1 as a documented convention — separate, and mostly docs (Adopt SimpleCov's run-visibility model: staleness, an env convention, and an off-vs-broken signal #276).
• .snap_diff_last_run.json and the staleness line — the bigger SimpleCov idea, tracked in Adopt SimpleCov's run-visibility model: staleness, an env convention, and an off-vs-broken signal #276.

Doing this one alone is still worth it: it stops training people to ignore the summary line.