fix: DagID schema pattern only start-anchored, breaks llama.cpp tool JSON schema validation

[LeXwDeX]

Why

llama.cpp's OpenAI-compatible server rejects any tool JSON Schema whose pattern is not anchored at both ends (^...$). The workflow tool's workflow_id parameter is branded as Dag.ID, whose Effect Schema filter Schema.isStartsWith("dag") serializes to pattern: "^dag" — start-anchored only. Every request carrying the workflow tool is rejected with JSON schema conversion failed: Pattern must start with '^' and end with '$', so an opencode session pointed at a llama.cpp backend fails on the first message.

Evidence: mitm capture of the tool payloads showed 5 unanchored ^dag patterns (one per workflow-tool parameter referencing Dag.ID); server log shows the corresponding AI_APICallError.

Scope

• packages/schema/src/dag-event.ts — make DagID's filter serialize to a fully anchored pattern (e.g. ^dag.*$) with identical validation semantics (starts-with).
• Regression test asserting the JSON Schema serialization of Dag.ID is both-start-and-end anchored.

Acceptance

• Generated tool parameter schemas for Dag.ID produce a pattern matching ^...\$ form accepted by llama.cpp (both anchors present).
• Existing DagID validation semantics unchanged (any string starting with dag_ still valid).
• specgit finish exits 0.