Improves performances

[arcanis]

• The original code was computing the join call even when the path didn't use either __virtual__ or .zip.
• Vectors were allocated empty. I updated that to preallocate a small arbitrary amount of items.

[github-actions]

Benchmark for ccf4c8b

Click to view benchmark

Test	Base	PR	%
vpath_edge_cases	3.0±0.10µs	2.5±0.08µs	-16.67%
vpath_mixed_workload	5.1±0.17µs	4.5±0.16µs	-11.76%
vpath_native	1909.5±62.57ns	1492.9±59.53ns	-21.82%
vpath_virtual	3.5±0.13µs	3.4±0.08µs	-2.86%
vpath_virtual_zip	2.9±0.12µs	2.8±0.08µs	-3.45%
vpath_zip	2.0±0.07µs	1916.7±61.78ns	-4.16%