Fix threading

[peter-reinholdt]

Some fixes for threading:

Previously, the chunk boundary was computed as:

ceil(sqrt(thread_idx / num_threads) * sideLength);

Since thread_idx / num_threads performed integer division, this always evaluated to 0 for thread_idx < num_threads.
I also removed the sqrt, since (as far as I can see), where the end_chunk_idx is used, it covers "linear" work, not triangular/quadratically growing work.

Additionally, I encountered segfaults on large problem sizes (many determinants), which I eventually tracked down to a heap-use-after-free from pyci/src/hci.cpp:277.
Not entirely sure why this would happen, but maybe the order of threads joining is not guaranteed to correspond to the order of the v_wfns? The proposed change first joins all threads, then adds the determinants:

    for (auto &thread : v_threads) thread.join();
    for (auto &wf : v_wfns) wfn.add_dets_from_wfn(wf);

[msricher]

Thank you!! I'll merge this soon.