Subcommands only work from the exact launch directory (path-hashed container names)

[alanbem]

Problem

get_container_name() hashes $PWD, and every subcommand (stop, rm, exec, ssh server, update, git, aws, gh, attach) resolves the container from the current directory. Consequences:

cd ~/projects/myapp && dclaude        # creates container A
cd ~/projects/myapp/src
dclaude stop                          # "No container found for this directory"
dclaude                               # silently creates a SECOND container for the same project

• Day-to-day commands fail confusingly from subdirectories
• Plain dclaude from a subdirectory spawns a duplicate container (own volume mounts, own SSH port) for what the user thinks is the same project
• Containers accumulate per-path with no dclaude ls to see them (cleanup requires the raw docker ps --filter name=dclaude incantation)

Proposed fix

1. Resolve the "project root" before hashing — walk up to the nearest .dclaude file or .git directory (the .dclaude walk already exists in find_config_file()), falling back to $PWD
2. Add dclaude ls to enumerate dclaude containers with their paths (path could be stored as a label at creation, e.g. dclaude.path=$HOST_PATH)

Backward-compat note: changing the hash input orphans existing containers; a transition warning or fallback lookup by label would help.

---

Found during a full-project code review.