perf: eliminate per-iteration allocation in fillBufferSingleBitAllocated

[suyashkumar]

Move the rawData buffer allocation outside the loop so a single []byte is reused across all iterations instead of allocating once per 8-pixel group. For a 512×512 1-bit image this reduces allocs from 32,770 to 3 per call and improves throughput by ~1.6× (379µs → 233µs).

Also adds BenchmarkFillBufferSingleBitAllocated to cover this path.

Metric	Before	After	Delta
512×512 time	379 µs/op	233 µs/op	−1.6× faster
512×512 allocs	32,770 allocs/op	3 allocs/op	−99.99%
512×512 memory	36,912 B/op	4,145 B/op	−88.8%
1024×1024 time	1,526 µs/op	933 µs/op	−1.6× faster
1024×1024 allocs	131,074 allocs/op	3 allocs/op	−99.99%
1024×1024 memory	135,216 B/op	4,145 B/op	−96.9%