Add performance benchmarks for large configurations

[jtdub]

Description

The project has pytest-profiling in dev dependencies but no dedicated benchmarks. For a library that processes network configurations — which can be thousands of lines on large devices — having baseline performance data would be valuable.

Areas where performance characteristics are unknown or potentially interesting:

• Parsing large configs (10,000+ lines) via _load_from_string_lines() vs get_hconfig_fast_load()
• config_to_get_to() on configs with many differences
• all_children_sorted() on deeply nested hierarchies (repeated sorted() calls)
• HConfigChildren.rebuild_mapping() cost on frequent deletions
• Memory usage for large config trees (benefit of __slots__)

Proposed Improvement

1. Add a benchmarks/ directory or pytest benchmark fixtures
2. Create representative large config samples (or generators)
3. Measure parsing, diffing, and iteration performance
4. Document expected performance characteristics in docs