[Feat] Client initialization cache and batch remove

[samzong]

What's changed?

• Add sync.Once-based lazy init to Client struct caching bareRoot, worktreeRoot, searchRoot, repoDir — resolved once per CLI invocation
• Add ListCached() / InvalidateList() for worktree list caching with explicit invalidation after mutations (Add, Remove, Dup, AddPR, Prune)
• Add RemoveBatch() with fail-fast validation, sequential removal, and batched branch -D
• Wire runWorktreeRemove in cmd layer to use RemoveBatch
• Refactor all public methods (Sync, Prune, Promote, AddPR, etc.) to use cached fields instead of per-call resolution
• Return defensive copy from ListCached() to prevent cache corruption
• Extract resolveRemoveTarget helper shared by prepareRemove and RemoveBatch

Why

• Client methods repeatedly compute values (FindBareRoot, GetWorktreeRoot, List()) that are constant within a single command lifecycle
• Measured ~45% reduction in redundant git subprocesses for batch operations (wt rm --all, wt add a b c, wt dup)
• RemoveBatch validates all names upfront (fail-fast for typos) and batches branch deletion at the end

[gemini-code-assist]

Code Review

This pull request introduces a caching mechanism for the git worktree list and centralizes repository initialization logic within the Client struct to improve performance and consistency. It also adds a RemoveBatch method for efficient multi-worktree removal and updates various operations to utilize the new cached state. Review feedback suggests further performance optimizations, specifically batching the git branch -D command and using internal synchronization methods to avoid redundant cache invalidations and git subprocess calls during worktree creation.