feat: shared squad — external state backend for multi-clone workflows #958

Problem

Some development workflows involve multiple clones of the same repository (worktrees, parallel clones, CI build environments, #multiple dev machines). In these environments:

• Branch switches can destroy untracked files including .squad/
• Multiple checkouts of the same repo each need their own squad init - duplicating team identity
• Team identity, decisions, and agent knowledge can't be shared across clones
• .squad/ committed to the repo isn't viable for large repos with strict policies on untracked content
• In very large monorepos, teams often work in isolated subtrees — committing .squad/ at the repo root may not be practical or permitted

Proposed Solution

Enable squad team state to live outside the repository — in a user-managed directory (git-backed squad repo or platform app data directory). Discovery happens via origin URL matching against a lightweight registry at ~/.squad/squad-repos.json.

Key design points

• Zero repo writes — shared mode writes nothing to the target repository
• URL-based matching — git remote get-url origin matched against registry patterns (GitHub, ADO, SSH URL forms all normalized)
• Git-backed squad repos — team state lives in its own repo, committable and shareable across machines
• Per-clone runtime state — clone-local ephemeral data stored under the platform's local app data directory (Windows/macOS/Linux)
• Cross-platform — ~/.squad/ pointer on all platforms; local data uses platform-appropriate paths

New commands

• squad init --shared — create or connect to a shared squad
• squad shared status|diagnose|list|doctor — operational tooling
• squad migrate --to shared — move existing squad state externally

Related

Partially addresses #943 (monorepo/persistence concerns)