fix(deployment): untangle execution + domain layers, fix strategy bug

[ntlaletsi70]

What

Deletes two confirmed-dead, unreconciled attempts at Deployment's Provider-dispatch design and a third unreconciled domain result/error model, fixes a real strategy-dispatch bug in the surviving path, and promotes pkg/intent/deployment to pkg/apis/deployment/intent to match the per-CR floor layout the rest of the codebase uses (pkg/apis/packages/intent/ already does this).

Why

Audited pkg/apis/deployment/ file-by-file after it was flagged as cluttered. Found three separate historical attempts at "dispatch a Provider from an Intent," only one of which NewDeploymentService actually constructs, plus a second domain model (Result/DeploymentState/ServiceUnitState/DeploymentError) with zero references anywhere outside its own three files. Every deletion below was verified via full diff or repo-wide grep before removal — nothing here had existing test coverage to break (find pkg/apis/deployment -name "*_test.go" was empty before this PR).

Real bug found and fixed along the way: K8SProvider.Execute (pkg/apis/deployment/api/kubernetes.go) had a parameter named intent, shadowing the intent package:

switch intent.Strategy {
case intent.Strategy:       // meant intent.StrategyRolling
    return p.executeRolling(ctx, intent)
case intent.Strategy:       // meant intent.StrategyBlueGreen — same bug
    return p.executeBlueGreen(ctx, intent)

Both case labels resolved to the parameter's own field, not the package constants — switch X { case X: } — so the first case always matched and executeBlueGreen was unreachable. Every Kubernetes deployment silently executed as Rolling regardless of configured Strategy. This compiles cleanly since it's not a constant comparison, so Go's duplicate-case check never caught it.

Domain

• environments
• events
• sources
• networks
• common
• Application / domain layer (no API surface change)
• CI / tooling / docs

API impact

• No API surface change — internal package restructuring, dead-code removal, and a bug fix restoring intended behavior

Checklist

• mage verify passes locally — no Magefile present in this checkout; ran go build ./..., go vet ./..., gofmt -l, and go test ./... instead — all clean, zero failures repo-wide
• buf breaking reviewed — N/A, no .proto changes
• Panic-free resolution — no panic() calls added; deleted code that used to panic
• Import paths use gen/go/blanketops/... for contract types — N/A, no contract import changes
• BlanketOps labels present where required — N/A
• Conditions written via core.SetCondition at each domain pipeline stage — N/A, no conditions logic changed
• Events emitted via core.EventRecorder for terminal outcomes — N/A
• ESP-0001 updated if contract semantics changed — N/A, no contract semantics changed
• Commit messages follow Conventional Commits

Notes for reviewer

• Detailed per-file evidence (which design is dead, which is live, and why) is in the commit message — every deletion is backed by either a full diff comparison or a confirmed-zero-references grep, not a guess.
• api/provider.go's ProviderRegistry becomes unused once application/backend_selector.go is deleted (its only caller), but is left in place — it's small, self-contained, reasonable plumbing for whenever Knative/ECS providers get built out, and isn't confusingly duplicated (only one ProviderRegistry exists, unlike the two BackendSelectors this PR removes down to one).
• Knative/ECS runtime support and real Canary/BlueGreen behavior remain unimplemented — acknowledged, pre-existing, explicitly out of scope for this pass.
• Every caller of the moved intent package dropped its explicit intent import alias, since the package's real name now matches (previously it was package deployment, colliding with sibling deployment packages, hence the alias).

🤖 Generated with Claude Code