⚡ use smallvec for better performance

[jvdd]

unsure if this will improve the performance, but seems worth exploring 👀

I believe we could use smallvec to write the output (usize array) of the downsampling methdos to. Currently we use Array1, but using smallvec could result in a performance boost 🙏

TODOs

• explore if collecting into vec is more efficient than pre-allocating and writing (is certainly more efficient when dealing with Mutexes, see 🕵️ parallel MinMaxDownsampler with x does not scale well to large n_out #14)
• explore the use of smallvec for this

---

Credits to this PR, from which I borrowed this idea!