refactor: align `gaze crap` threshold model with `gaze report` (`*int` + `Flags.Changed`)

[yvonnedevlinrh]

• Severity: Low
• Status: Confirmed - documented in PR fix: prevent silent CI passes on missing/partial data (#101, #108, #116) #148 design artifact
• Component: cmd/gaze/main.go - checkCIThresholds (lines 686-691)
• Affected command: gaze crap
• Related: Bug: report --max-gaze-crapload always PASSES when GazeCRAP is unavailable #108, PR fix: prevent silent CI passes on missing/partial data (#101, #108, #116) #148

Summary

gaze crap and gaze report use different threshold models for
--max-gaze-crapload, producing different exit codes for the same
unavailable-data condition:

• gaze report (EvaluateThresholds): *int thresholds +
  cmd.Flags().Changed(). When the threshold is set but GazeCRAPload is
  nil, it fails with "GazeCRAPload (unavailable)".
• gaze crap (checkCIThresholds): plain int thresholds with
  0 = disabled. When GazeCRAPload is nil, it silently skips the check.

Reproduction

# gaze report: threshold set, GazeCRAP unavailable - FAIL
$ gaze report --format=json --max-gaze-crapload=5 . 2>&1 | grep GazeCRAPload
GazeCRAPload (unavailable): N/A/5 (FAIL)

# gaze crap: threshold set, GazeCRAP unavailable - silently skipped
$ gaze crap --max-gaze-crapload=5 . 2>&1 | grep -i gaze
note: GazeCRAP unavailable - run 'gaze quality' to compute contract coverage
$ echo $?
0

Expected vs Actual

• Expected: Both commands should produce the same pass/fail result for the
  same threshold flag and the same unavailable-data condition.
• Actual: gaze report fails the gate; gaze crap silently skips it.

Root cause

checkCIThresholds in cmd/gaze/main.go uses plain int flags with
0 = no limit semantics. The nil-check at line 689 skips the comparison
when GazeCRAPload is nil, but does not fail the gate:

if maxGazeCrapload > 0 && rpt.Summary.GazeCRAPload != nil &&
    *rpt.Summary.GazeCRAPload > maxGazeCrapload {

This also means --max-crapload=0 is treated as "disabled" rather than
"fail if any function exceeds 0" - a second divergence from
gaze report's *int model where zero is a live threshold.

Origin

The ci-gate-integrity design document (PR #148,
openspec/changes/ci-gate-integrity/design.md) lists alignment as a
goal but explicitly defers the refactoring as a non-goal:

Aligning gaze crap's int-based threshold model with gaze report's
*int + cmd.Flags().Changed() model - that's a separate refactoring
task.

The code comment at cmd/gaze/main.go:686-691 also documents the
divergence.

Suggested fix

Refactor gaze crap threshold handling to use *int flags and
cmd.Flags().Changed() for all threshold checks (not just the zero-result
gate added in #148). Unify checkCIThresholds with EvaluateThresholds
or replace it entirely so both commands share the same threshold semantics.