fix: always emit non-empty deploy_name in notify dispatch

[joshua-temple]

Problem

A build-only satellite (manifest with builds + a notify to a primary repo, but zero deploys) generated a "Notify Primary Repo" step whose dispatch to the primary's external-update.yaml omitted the deploy_name input. The consumer declares deploy_name as required: true, so GitHub rejected the dispatch with 422 Required input 'deploy_name' not provided (observed live on cascade-example-artifact-a and artifact-b).

Root cause: writeNotifyPrimaryStep emitted deploy_name only if len(g.config.Deploys) > 0.

A latent typo in the same step shared the same root cause: the github-script accessor was written as context.jobs.['id'] with a stray dot before the bracket (invalid JavaScript). It only renders for satellites that collect deploy outputs, so it was dormant for zero-deploy manifests but would break a notify+deploys satellite.

Fix

• deploy_name is now always emitted and non-empty. Fallback chain: first deploy name, then first build name (the build-only artifact case), then the satellite's own repo name from the runtime context (context.repo.repo). The chosen value identifies the satellite's artifact and matches the --deploy-name required flag on cascade external update.
• Stray dot removed: context.jobs.['id'] -> context.jobs['id'].

Verification

• Failing-first tests added in internal/generate/external_test.go: a build-only builds+notify manifest asserts the Notify step emits a non-empty deploy_name; a notify+deploys manifest asserts the accessor has no stray dot.
• go build ./..., go test ./... (1395 pass), golangci-lint run ./... all green.
• Dogfood regen produces zero workflow drift (cascade's own manifest has no notify block).