Reduce allocations during parsing

[diegommm]

Current parsing (and lexing) is very efficient in terms of direct CPU and memory usage, but it has an indirect cost on CPU because it performs many little allocations, which increases GC pressure.
In use cases where a large number of expressions need to be parsed, this can cause spikes in allocations whereas it can be rather easy to avoid.

I will provide a PR with improvements and benchmarks.