feat(data ingest): expected image size + local min-size floor preview (#183)

[LukasWodka]

Summary

Two image-preflight improvements for tracebloc data ingest, so a customer finds out about size problems before the ingest, not after (their data never leaves the box either way):

• Expected image size — --target-size is reframed as a check, not a resize. tracebloc validates every image is exactly that size and rejects any that differ; the CLI previews the same reject locally. Default: read from your first image.
• Min-size floor preview — --min-size WxH previews data-ingestors' minimum-image-size floor (ci: bump GitHub Actions (bundles #342–#346, matches artifact majors) #348) locally. It mirrors the ingestor's _meets_min_size exactly: too small when either side is below the floor, boundary inclusive, [W, H] order, no swap. The override travels as spec.file_options.min_size.

Closes#183.

What changed

• internal/cli/data.go — --target-size / --min-size flags + help; plumb --min-size to spec.file_options.min_size.
• internal/push/spec.go — SpecArgs.MinSize, MinImageSize constant (the mirror of the upstream default), emit under file_options.
• internal/push/preflight.go — ValidateImages previews the floor (floor-first, then the resolution mismatch, matching the ingestor's error order); PreflightDataset wiring.
• internal/schema/ingest.v1.json — re-synced to the pinned ref efaeb07, which learned about min_size.
• Parity harness (parity_golden_test.go, scripts/gen-validator-goldens.py, testdata/parity/) — min_size is now representable on both sides, with a new end-to-end case.

Review follow-ups folded in (last commit)

• Deployment parity. The floor exists in data-ingestors only on develop (di#348/fix(doctor): requests-proxy check — result-egress framing + honest ✔ (cli#351 Part 1) #356); the deployed ingestor (v0.5.7/v0.6.0) has no floor and ingests small images fine. So the preview no longer applies the 32x32 default on its own — that would exit-3 an ingest the live cluster accepts, the inverse of the tabular-BOM block (whose reject mirrors a real deployed rejection). The floor is previewed only when the customer sets --min-size (their own declared requirement). imgc-too-small is now a flagged deliberate divergence.
• Message. The too-small reject no longer claims "the cluster rejects images below this size" — false for the deployed cluster. It speaks to the floor the customer set.
• Flag help.--min-size no longer calls 32x32 "the smallest tracebloc will ingest" (which contradicted both the reject's "lower the floor with --min-size" and the feature — 16x16 is accepted end-to-end). Reworded to raise-or-lower, honest "no local size check" default.
• Parity gap.min_size was representable on neither side of the harness, so the headline override was cross-checked by neither. Added imgc-min-size-override: 24x40 images with an explicit [16,32] floor that the default 32x32 would reject — pinning that the override is actually read (not silently the default) and its [W, H] orientation is honored, on both the CLI and the real ingestor.

Test plan

Local, all green:

• go build ./...
• go test -race -cover ./... (incl. TestValidatorParity, TestValidateImagesMinSize, TestBuild_WithMinSize_PassesSchema)
• gofmt -l . / gofmt -s -l . → empty
• go vet ./...
• ./scripts/sync-schema.sh --check → no drift (pinned ref efaeb07)
• ./scripts/coverage-floor.sh → floors met
• Goldens regenerated from the pinned ref: only imgc-min-size-override added; every existing golden byte-stable (incl. imgc-too-small → reject).

Decisions to ratify

1. Floor value mirrors di#348 exactly — MinImageSize = {32, 32}, either-side / boundary-inclusive / [W, H]. It is the CLI's mirror of ImageResolutionValidator.MIN_IMAGE_SIZE; do not invent a different floor here (Principle 6).
2. Floor is opt-in until di#348 reaches prod. Today the preview enforces the floor only via explicit --min-size; the default 32x32 is not previewed, because the deployed ingestor has none. When di#348 ships to prod, default PreflightDataset to MinImageSize and flip imgc-too-small back to reject (both flagged inline + in cases.json).
3. Schema re-sync.internal/schema/ingest.v1.json is pinned to efaeb07 (via scripts/.data-ingestors-ref), the ref that added min_size. Adopting further upstream changes stays a deliberate SHA bump + re-sync.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches image preflight and ingest spec emission with deployment-parity caveats; behavior is heavily tested via parity goldens but wrong floor defaults could block or allow bad ingests until prod catches up with #348.

Overview
Adds --min-size WxH for image ingest tasks so customers can set a minimum resolution floor locally and in the emitted spec (spec.file_options.min_size), aligned with data-ingestors #348. --target-size help and interactive copy now stress that tracebloc validates size and never resizes.

Local preflight extends ValidateImages with an optional min-size check (either dimension below the floor fails; floor errors beat target-size mismatches). The preview applies that floor only when --min-size is set, not the upstream 32×32 default, so the CLI does not reject ingests that today’s deployed ingestor still accepts without a floor.

data ingest rejects --target-size / --min-size on non-image tasks at exit 2 instead of silently ignoring them. ParseMinSize shares WxH parsing with target size; SpecArgs.MinSize and schema re-sync (min_size on file_options, pinned ingestors ref) wire through Build. Parity fixtures and goldens were updated (including deliberate CLI vs pinned-ingestor divergences for default floor and tabular BOM).

^{Reviewed by Cursor Bugbot for commit fd4688e. Bugbot is set up for automated code reviews on this repo. Configure here.}