[hellerve]

this pr adds a bunch of performance improvements to core:

1. String.starts-with? and String.ends-with? now compare bytes. They used to decode the whole string into chars, and now compare bytes and stop at the first mismatch. This also fixes (starts-with? "ölheller" "öl"), which returned false. I measured 9.4µs → 0.1µs on a 4KB string.
2. Array.swap! now moves instead of copying, because it used to make two copies and two deletes per swap. Heapsort (one of the consumers) on 2k strings: 1.22ms → 0.36ms
3. Map and Set now use open addressing, which was on our backlog for a while. Uses linear probing over a column of stored hashes, Fibonacci hashing, and removal without tombstones. Also, in-place operations grow the table, and the functional versions stop copying it. 10k int put!: 3.2ms → 59µs
4. Semi-related: I moved Map.update-value! / Map.value-ref! into core from web,. They pass a reference to the stored value without copying it.
5. String.split-by, slice, prefix, suffix/ split-by ran strlen on the whole string once per separator, so lines and words were accidentally quadratic. Also, slice decoded the whole string. Both are now single byte scans. lines on 90KB went from 18ms to 0.25ms.
6. Array.map-reduce, slice, remove-nth: map-reduce copied its result array on every step. slice grew its result per element. remove-nth copied every element after nth. 10k map-reduce: 2.86ms → ~1µs
7. For str on Map, Set, and Bag each entry used to copy the whole string built so far. They now collect the pieces and join them once. 1.68ms → 0.45ms on 2k-entry Map.str.

Added a whole bunch of tests.

Cheers

[hellerve]

Added one more: Map.value-ref! can be used for Map.=! Faster and cleaner.

[eriksvedang]

Very nice, thanks a lot for this!