feat(storyboard): add field_greater_than / field_at_most / field_at_least matchers (#1839)

[bokelley]

Summary

Closes #1839.

Adds two new storyboard validation check kinds:

• field_at_most — assert numeric field ≤ comparand (non-strict)
• field_at_least — assert numeric field ≥ comparand (non-strict)

They share the same comparand-resolution model as the existing strict field_less_than: either a literal value or a runtime-captured context_key, with absent context keys passing the check with a context_key_absent observation (the prior step may have been legitimately skipped on a branch-set path).

Why

Storyboard scenarios that assert "value X is at most / at least N" — cap and floor enforcement — have no clean primitive today. The strongest available is field_less_than (strict <), which forces a literal-plus-epsilon workaround:

- check: field_less_than
  path: media_buy_deliveries[0].totals.frequency
  value: 3.01  # cap is 3; epsilon to allow equality

That's semantically wrong (assumes a domain rounding convention) and brittle when sellers report the cap value exactly (3.00 < 3.01 passes; 3.00 < 3.00 fails).

field_at_most: 3 is the right primitive.

Implementation note

Refactored validateFieldLessThan into a shared validateNumericComparison(op) helper parameterised by check name, operator symbol, and comparator function. All three numeric checks dispatch through it — no copy-paste branches. Forward-compat default in the runner already grades unknown check kinds as not_applicable, so older runners won't brick.

Concrete consumer

media_buy_seller/frequency_cap_enforcement (adcontextprotocol/adcp#4727) ships with the epsilon workaround. Once this lands, a small follow-up will swap field_less_than: 3.01 → field_at_most: 3 in that scenario yaml.

Test plan

• 12 new unit tests covering both matchers (test/storyboard-cross-step-validators.test.js):
  • strictly-less / strictly-greater paths
  • boundary-equality paths (the cases that distinguish ≤/≥ from </>)
  • literal-value comparand
  • context_key comparand
  • context_key_absent observation
  • type-error: missing field
  • type-error: non-numeric field
  • type-error: non-numeric comparand
  • missing path
• All 22 existing field_less_than / field_equals_context tests still pass (refactor preserves behaviour).
• npm run format:check clean.
• tsc --noEmit clean.

Python parity

Per the issue: once TS ships, mirror in adcp-client-python. Not addressed here.

🤖 Generated with Claude Code

[bokelley]

Review pass — done

Parallel reviews from code-reviewer + ad-tech-protocol-expert.

code-reviewer: ship. Refactor preserves all original behavior (error messages, expected/actual fields, observation pass-through, type-error paths) exactly; boundary semantics correct; NUMERIC_OPS typing sound (as const satisfies …).

ad-tech-protocol-expert: sound-with-one-caveat. Naming matches the prose-style vocabulary (field_present, array_length, contains) rather than operator-symbol style (field_lte) — right call. Forward-compat verified. Flagged: shipping ≤/≥ without > leaves the comparator quadrant 3-of-4 filled. Storyboard authors reaching for strict-> would hit the runner's forward-compat not_applicable default and silently grade as passing — semantically wrong.

Addressed in ba4faed4: added field_greater_than to close the quadrant. One more NUMERIC_OPS entry; the shared validateNumericComparison helper handles it without duplication. Tests added covering the strict-> equality-boundary case (the unique distinguishing behavior vs field_at_least). 38/38 tests pass.

Title updated to reflect all three matchers.