feat: Enhanced Tab Completion for Interactive Mode

## Overview
Implement advanced tab completion capabilities for bssh's interactive mode, including remote path completion, command completion based on remote $PATH, and history-based completion suggestions. This will significantly improve the user experience by providing context-aware autocompletion similar to modern shells.

## Technical Approach

### Architecture Design
The implementation will use rustyline's built-in completion framework with custom completers that leverage SSH channels to fetch remote information. We'll implement a multi-tiered completion system:

1. **Completion Engine Architecture**
- Create a composite completer that delegates to specialized completers based on context
- Cache remote data to minimize SSH round-trips
- Implement async completion fetching with timeout handling
- Use a priority system: history > remote paths > remote commands

2. **Data Caching Strategy**
- Cache remote $PATH directories and executables for 5 minutes
- Cache recently accessed directory listings for 1 minute
- Maintain a sliding window of command history (last 1000 commands)
- Implement cache invalidation on directory changes

3. **Context Detection**
- Parse current command line to determine completion context
- Detect if completing a command, path, or argument
- Handle special characters and escape sequences
- Support both absolute and relative path completion

## Key Features

- [ ] Tab completion for remote paths with <100ms latency (cached)
- [ ] Command completion includes all executables in remote $PATH
- [ ] History-based suggestions appear within 50ms
- [ ] Completion works correctly across node switches
- [ ] Cache persists between sessions

## Implementation Phases

### Phase 1: Foundation and Architecture
- [ ] Create completion module structure at `src/completion/mod.rs`
- [ ] Implement context parser at `src/completion/parser.rs`
- [ ] Create cache manager at `src/completion/cache.rs`
- [ ] Implement SSH data fetcher at `src/completion/remote_fetcher.rs`

### Phase 2: Core Completion Implementations
- [ ] Implement remote path completer at `src/completion/path_completer.rs`
- [ ] Implement remote command completer at `src/completion/command_completer.rs`
- [ ] Implement history-based completer at `src/completion/history_completer.rs`
- [ ] Create composite completer at `src/completion/composite_completer.rs`

### Phase 3: Integration with Interactive Mode
- [ ] Extend `InteractiveCommand` to use custom completer
- [ ] Add completion configuration to `InteractiveConfig`
- [ ] Implement async completion updates
- [ ] Add multi-node completion support

### Phase 4: Advanced Features and Optimization
- [ ] Implement smart completion features (git, docker, kubectl argument completion)
- [ ] Add visual completion enhancements (color-coding, hints, preview)
- [ ] Optimize performance (incremental updates, bloom filters, batching)
- [ ] Add completion debugging and metrics

### Phase 5: Testing & Polish
- [ ] Write comprehensive unit tests
- [ ] Add integration tests
- [ ] Update documentation
- [ ] Polish user experience

## Dependencies

### External Dependencies
- `rustyline = "17.0"` - Already present, provides completion framework
- Consider adding `fuzzy-matcher = "0.3"` for fuzzy history matching
- Consider adding `lru = "0.12"` for LRU cache implementation
- May need `shellexpand = "3.1"` for path expansion

### Internal Dependencies
- Existing SSH client implementation (`src/ssh/tokio_client/`)
- Existing interactive mode infrastructure (`src/commands/interactive.rs`)
- Configuration system (`src/config.rs`)
- Node session management

## Success Criteria

### Functional Requirements
- Tab completion works for remote paths with <100ms latency (cached)
- Command completion includes all executables in remote $PATH
- History-based suggestions appear within 50ms
- Completion works correctly across node switches
- Cache persists between sessions

### Performance Benchmarks
- Initial PATH scan completes within 2 seconds
- Subsequent completions served from cache in <10ms
- Memory usage for caches stays under 50MB
- No blocking of interactive input during completion
- Support for directories with 10,000+ files

### User Experience
- Completion feels as responsive as local shell
- Clear visual feedback for completion types
- Intuitive behavior matching user expectations
- Graceful handling of network issues
- Consistent behavior across all nodes

## Related Work
- Original task document: `.claude/tasks/todo/1_enhanced_tab_completion.md`