gc prime: add --strict flag to error on unknown agents

[csells]

Summary

gc prime <name> silently falls back to a generic default prompt when the agent doesn't exist. This is useful as a default behavior — any agent without a custom prompt gets a reasonable starting point. But when debugging, there's no way to distinguish "my agent's actual prompt" from "the fallback because I typo'd the name."

Example

$ gc prime dfdfdf
# Gas City Agent

You are an agent in a Gas City workspace...

No error, no warning. If you're trying to verify your agent's prompt template renders correctly, this is a footgun.

Proposal

Add a --strict flag that errors if the agent isn't found in the config:

$ gc prime dfdfdf --strict
Error: agent "dfdfdf" not found in city config

Without --strict, current behavior is preserved — the default prompt is returned for any agent.

[csells]

Additional: gc prime should populate all template variables for debugging

Currently buildPrimeContext() only populates WorkDir, RigName, RigRoot, and Branch from GC_* environment variables — which are only set inside managed sessions. When run from the terminal for debugging, these render as empty strings (thanks to missingkey=zero), giving a misleading preview.

For example, {{.WorkDir}} silently disappears from the output, making it look like the template is broken when it would work fine at runtime.

Recommendation: When GC_* env vars aren't set, gc prime should fall back to reasonable values:

Variable	Fallback
WorkDir	Agent's configured dir, or the rig root, or cwd
RigName	Resolved from agent's rig association
AgentName	Already handled (falls back to a.QualifiedName())
Branch	Current git branch
DefaultBranch	Already handled via defaultBranchFor() but gets empty string if WorkDir is empty

An empty variable in gc prime output is never helpful — it should always show what the agent would actually see at runtime.