fix: delegate on-demand pricing to truffle instead of fabricating a rate (#533)

[scttfrdmn]

Summary

• --cost-limit was enforced against a fabricated price, not the real on-demand rate. At launch, when PricePerHour wasn't already known, spawn called its own hand-rolled, uncached AWS Pricing API lookup (LookupEC2OnDemandPrice) and, on failure — which it always did for any principal onboarded via spawn onboard, since that policy granted no pricing:* action — fell back to a static per-family-size guess table (libs/pricing.GetEC2HourlyRate) missing every current-gen instance family (g6, g6e, p5, p5e, c7g, c8g, ...). Measured: g6e.xlarge guessed at $0.20/hr vs a real $1.861/hr (9.3x under); p5.48xlarge guessed at $9.60/hr vs a real $55.04/hr (5.7x under). That fabricated number is what gets written to spawn:price-per-hour and is what spored enforces --cost-limit against.
• Pricing at launch now delegates entirely to truffle's OnDemandPriceWithSource (github.com/spore-host/truffle/pkg/aws) — the suite's pricing authority, with a real 24h cache and a static fallback that errors rather than guesses (truffle#114/test(agent): guard --on-complete fires regardless of job origin (#105) #115). spawn keeps no rate table of its own for enforcement anymore.
• If truffle can't price the instance at all (neither live nor its own static table) and the launch requested --cost-limit, the launch now fails with an error naming the instance type, region, and limit — instead of silently leaving the cap unenforced. Without --cost-limit, an unpriceable instance still launches with PricePerHour left unset (spored's cap-check already treats 0 as "no cap enforced," never fabricated).
• Removed spawn's own LookupEC2OnDemandPrice (dead code after the above). libs/pricing.GetEC2HourlyRate is intentionally left in place for the separate display-only estimate paths (--estimate-only's "On-demand:" line in cmd/launch_single.go, spawn cost in cmd/service.go/pkg/cost) — that's a companion display bug tracked separately, out of scope here per the issue.
• Added pricing:GetProducts to spawn onboard's IAM policy (cmd/onboard.go) — a global, read-only action with no resource scoping — since the pricing lookup runs in-process under the launching principal's own credentials via truffle. Operational note: anyone onboarded before this change needs to re-run spawn onboard to pick this up.
• Bumped github.com/spore-host/truffle v0.49.0 → v0.53.0. Diffed pkg/aws/pricing.go between the two versions — byte-identical, no signature changes to OnDemandPrice/OnDemandPriceWithSource/GetCapabilities. Skimmed truffle's CHANGELOG for v0.50.0–v0.53.0: additive fields (Capabilities), new discovery-path error-handling fixes, find/search UX changes — nothing affecting the call sites spawn already uses (cmd/slurm.go, cmd/task.go, cmd/launch_single.go, cmd/launch_sweep_quota.go, cmd/launch_preflight.go).
• cmd/status.go's worst-case-cost display now reads the spawn:price-per-hour tag (via the existing pricePerHour helper) instead of independently re-deriving a rate, so the displayed ceiling can't drift from what --cost-limit is actually enforced against.

Notes on searches performed

• Grepped for the "fabrication-asserting" test the issue expected around pricing_test.go:69-70 (unknown.xlarge → expected: 0.2). That test lives in the separate github.com/spore-host/libs module's pkg/pricing/pricing_test.go (a different repo/module), not in spawn — spawn itself had no pricing_test.go before this PR. Nothing to delete in this repo; libs/pricing.GetEC2HourlyRate and its test are untouched, since that package is still legitimately used for the display-only estimate paths noted above.

Test plan

• New unit tests in pkg/aws/pricing_test.go (TestResolvePricePerHour_*) mock the truffle pricing call via a small onDemandPricer interface (no real AWS calls):
  • a successful lookup (live or truffle's own static source) populates PricePerHour with the real rate
  • a failed lookup + --cost-limit requested → error naming instance type/region/limit
  • a failed lookup + no --cost-limit → launch proceeds, PricePerHour stays 0
  • a caller-supplied PricePerHour is never overridden
• Fail-without/pass-with verified by mutation: temporarily broke the cost-limit gate (cfg.CostLimit > 0 → cfg.CostLimit > 999999), confirmed 2 of the new tests failed, reverted, confirmed all pass.
• make check clean (fmt, vet, staticcheck skip — not installed locally —, full short test suite across all packages).
• go build ./... and go vet ./... clean after the truffle bump.

[codecov]

Codecov Report

❌ Patch coverage is 75.00000% with 6 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
pkg/aws/client.go	0.00%	2 Missing ⚠️
pkg/aws/pricing.go	90.00%	2 Missing ⚠️
cmd/onboard.go	0.00%	1 Missing ⚠️
cmd/status.go	0.00%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!