Optimize hierarchy building and metadata folding performance

[mountainMath]

Summary

• Optimize add_hierarchy() function with memoized path tracing instead of iterative while loop
• Optimize fold_in_metadata_for_columns() with pre-split coordinate matrix
• Optimize factor conversion with lazy matrix initialization

Performance Benchmarks

add_hierarchy() - Hierarchy Building

Test Case	Original	Optimized	Improvement
6 small dimensions (2-11 members each)	0.0243s	0.0058s	76% faster
NAICS hierarchy (337 members)	0.0314s	0.0037s	8.5x faster (88%)

The optimization replaces an iterative while loop (up to 100 iterations) with single-pass memoized path tracing:

• Old: Each iteration does strsplit() on all rows to find current top, paste0() to prepend parent, string comparison to check changes
• New: Build parent lookup once, trace each node to root with cached ancestor paths

fold_in_metadata_for_columns() - Coordinate Parsing

Test Case	Original	Optimized	Improvement
169k rows, 6 dimensions	0.057s	0.059s	~same

The matrix-based approach has similar overhead to lapply for typical table sizes.

Correctness Verification

Verified exact output equivalence by installing both package versions and comparing results:

Table 20-10-0085 (169,108 rows, 33 columns, 6 hierarchy columns):

Column names identical: TRUE
Data identical: TRUE
All hierarchy columns identical: TRUE

Table 20-10-0056 (73,530 rows, 27 columns, 4 hierarchy columns):

Column names identical: TRUE  
Data identical: TRUE
All hierarchy columns identical: TRUE

Test plan

• All existing tests pass (devtools::test() - 20 tests)
• devtools::check() passes with 0 errors, 0 warnings
• Output verified identical on tables 20-10-0085 and 20-10-0056
• Hierarchy columns specifically verified identical

🤖 Generated with Claude Code