perf(csv-parse): avoid unnecessary allocation in ResizeableBuffer.toString

[CamWass]

This PR optimises ResizeableBuffer.toString in csv-parse by avoiding an intermediate sliced Buffer created by this.buf.slice(...) and instead passes the bounds directly to Buffer.toString.

Although this allocation is a small cost, it adds up given this method is called for each field in the input.

Benchmarks

Existing benchmarks

Take these with a grain of salt - the existing benchmark suite only parses each input file once, making them quite noisy (especially for the smaller inputs)

Length	Execution Time (Before)	Execution Time (After)	Throughput (Before)	Throughput (After)	Throughput Improvement
20,000	0.622 s	0.356 s	32,131 op/s	56,129 op/s	+74.69%
200,000	1.563 s	1.256 s	127,969 op/s	159,265 op/s	+24.46%
2,000,000	8.498 s	7.212 s	235,339 op/s	277,296 op/s	+17.83%
20,000,000	44.022 s	36.789 s	454,316 op/s	543,641 op/s	+19.66%

Custom benchmark

To get a more accurate benchmark, I modified the existing benchmark to only test a input file of length 200,000, and took the average of 100 iterations:
0.386769533s before -> 0.340958805s after = ~12% improvement