feat(v2): add Parse() function and extensible Registry for model string resolution

Introduces llm.Parse(spec) backed by an extensible Registry that resolves
model strings like "openai/gpt-4o", aliases like "fast", and named targets
like "m5/qwen3:30b" (via LLM_M5 env var DSNs) into ready-to-use *Model
objects. Extension points: RegisterProvider, RegisterAlias, RegisterResolver.
Adds Foreman constructor and sentinel errors ErrAliasLoop, ErrUnknownProvider,
ErrInvalidDSN.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

v2/errors.go

@@ -17,4 +17,16 @@ var (
 
 	// ErrNoStructuredOutput is returned when the model did not return a structured output tool call.
 	ErrNoStructuredOutput = errors.New("model did not return structured output")
+
+	// ErrAliasLoop is returned when alias resolution exceeds the maximum depth (10),
+	// indicating a cycle such as "a" → "b" → "a".
+	ErrAliasLoop = errors.New("alias resolution loop detected (depth > 10)")
+
+	// ErrUnknownProvider is returned when a spec references a provider name that
+	// is not registered and has no corresponding LLM_X environment variable.
+	ErrUnknownProvider = errors.New("unknown provider")
+
+	// ErrInvalidDSN is returned when a DSN string (from an LLM_X env var) cannot
+	// be parsed. Expected format: scheme://[token@]host[/path].
+	ErrInvalidDSN = errors.New("invalid DSN")
 )