v0.1.7 — Fastest SPSS Reader

Performance: 6 Optimizations

ambers is now faster than polars_readstat on every file, from small (0.2 MB) to large (1.1 GB).

Optimizations

1. Bytecode match reorder — hot path 1..=251 first for better branch prediction
2. Cow<str> for string decoding — eliminates heap allocation for UTF-8 files (most modern SPSS)
3. Bulk I/O + buffer reuse — single read_exact per row for uncompressed data (was 72M individual reads on 1.1 GB file)
4. Pre-computed VLS segment layout — avoids redundant per-row arithmetic for very long strings
5. Smart capacity hints — StringBuilder uses actual variable width instead of hardcoded 32 bytes
6. StringViewArray with deduplication — Utf8View + deduplicate_strings() for categorical SPSS data

Results (avg 5 runs, Polars DataFrame output)

File	ambers	polars_readstat	pyreadstat	vs polars_readstat	vs pyreadstat
test_1 (0.2 MB, bytecode)	0.002s	0.004s	0.328s	2.0x faster	175x faster
test_2 (147 MB, bytecode)	0.880s	0.949s	3.618s	1.1x faster	4.1x faster
test_3 (1.1 GB, uncompressed)	1.094s	1.359s	5.002s	1.2x faster	4.6x faster
test_4 (0.6 MB, uncompressed)	0.013s	0.015s	0.022s	1.1x faster	1.7x faster
test_5 (0.6 MB, uncompressed)	0.002s	0.004s	0.016s	1.9x faster	8.2x faster

Lazy Pushdown

File	Full collect	Select 5 + head 1000	Speedup
test_2 (147 MB)	0.833s	0.084s	9.9x
test_3 (1.1 GB)	1.036s	0.006s	167x

On the 1.1 GB file, selecting 5 columns and 1000 rows completes in 6ms.

Full Changelog: v0.1.6...v0.1.7