refactor(sdk): reframe sei SDK as a wire-client product (Phase A)

[bdchatham]

What

Make the sdk/sei pieces idiomatically express a settled design decision: the SDK is a wire-client product + contract owner. Its providers (k8s, docker) are WIRE clients — they speak RPC/CRDs to an already-running chain, so version-pinning them is fine. An in-process ("local") provider is a category error for the SDK: it would link sei-chain's Go source (the controller can't transitively build it; and it co-versions the harness with the chain). The in-process harness lives in sei-chain, satisfies these handle interfaces by shape, and is registered/constructed consumer-side. The SDK stops pretending to host "local".

Changes

• Delete the sdk/sei/provider/local ErrNotImplemented stub package.
• Drop the built-in local mode + SEI_LOCAL from resolveMode (and the resolve/ambiguity error text). The registry stays generic — a consumer may still provider.Register("local", …) its own in-process adapter and Open(ctx, "local") explicitly.
• Keep docker untouched (legitimate wire stub).
• Name the contract seam: Provider/*Handle + *Spec + the registry are the stable, dependency-light contract — the seam a future leaf module would extract — and must stay free of controller-runtime/apimachinery/CRD imports. Documented in provider.go and CLAUDE.md. Not extracted now (deferred).
• Add a go list -deps import-discipline test that mechanically enforces that seam on the core sei package.
• Fix the drift test to pin built-in modes to exactly {k8s, docker}; refresh package/Provider/Object() docs and sdk/CLAUDE.md.

No Provider/handle interface signatures change. No contract sub-package/module extraction, no harness changes, no sei-chain changes (Phases B/C).

Verification

go build ./..., go test ./sdk/... -race, go vet ./sdk/..., gofmt -s -l sdk/, golangci-lint run ./sdk/... (0 issues) — all green.

For reviewers

• The drift test moved from "contains" to "exactly equals {docker, k8s}" — stricter, so an extra built-in mode also fails. Intentional: the SDK's built-in surface is now closed.
• TestResolveMode_ExplicitBeatsEnv now resolves a "consumer-mode" name, doubling as coverage that a consumer passes its own mode explicitly (the only path left for a non-wire provider).
• The new imports_test.go shells out to go list; it confirmed core has zero hits while the k8s provider pulls apimachinery, so it discriminates correctly.

Out of scope (Phase A)

Interface-signature changes; contract sub-package/module extraction; integration harness (Phase B); sei-chain (Phase C).

🤖 Generated with Claude Code

[cursor]

PR Summary

Medium Risk
Breaking for callers that relied on SEI_LOCAL or blank-importing the local stub; mode resolution and registered built-in providers change, though explicit Open and RegisterProvider remain for custom modes.

Overview
Repositions the SDK as a wire-client product (k8s/docker only) and drops the shipped local stub and SEI_LOCAL env wiring. In-process harnesses are expected to live in sei-chain and register via RegisterProvider with an explicit Open(ctx, "…") mode name.

Removals: entire sdk/sei/provider/local package (stub + tests); modeLocal and SEI_LOCAL from resolveMode and related error strings.

Documentation: sdk/CLAUDE.md, sei.go, provider.go, and readiness.go now describe the contract seam, consumer-side registration, and that Object() may be nil for non-k8s providers.

Tests: new TestCoreContractStaysDependencyLight (go list -deps) bans k8s/controller-runtime/api imports from core sdk/sei; TestProviderKeysMatchOpenModes requires built-in providers to be exactly {docker, k8s}; registry/Open tests updated for two env modes and explicit consumer mode names.

Public **Provider/handle signatures are unchanged.

^{Reviewed by Cursor Bugbot for commit ba8ccf7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[bdchatham]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit ba8ccf7. Configure here.}

[bdchatham]

Review-gate: PASS — merging

Coral slate (blinded, 4 lenses): unanimous RATIFY.

• kubernetes-specialist (assigned dissenter): RATIFY — repo-wide hunt found zero live consumers of the removed SEI_LOCAL/local surface; sole SDK consumer (integration harness) is k8s-only; imports_test seam guard discriminates correctly; SEI_LOCAL not on the frozen one-way-door list; provider.Register + explicit Open reintroduction path open.
• systems-engineer: RATIFY — import-guard fails-closed and can fire (verified via go list -deps); resolveMode correct across all cases; registry concurrency untouched; drift exact-equals sound.
• idiomatic-reviewer: RATIFY — reads native. Advisory: fold the duplicated seam comment into the Provider doc.
• prose-steward: RATIFY (w/ reservations) — docs unambiguous. Advisory: standardize on the provider.Register façade (vs RegisterProvider) in consumer-facing instructions.

Bugbot: clean. CI: lint / test / test-integration / verify-generated all green.

The two advisory doc nits are non-blocking and will be folded into a later SDK-docs touch.