Multithreading attempt

[lkdvos]

This is me playing around a bit with trying to multithread the commutator.
While it does gain a little bit over the optimized version of #25, the results aren't that impressive:

With 8 threads on my machine:

|                            | 9f209c0dbabaa5... | dirty           | 9f209c0dbabaa5.../dirty |
|:---------------------------|:-----------------:|:---------------:|:-----------------------:|
| lanczos (N=50)/nterms=2^14 | 0.473 ± 0.019 s   | 0.583 ± 0.019 s | 0.81                    |
| lanczos (N=50)/nterms=2^16 | 1.4 ± 0.041 s     | 1.33 ± 0.017 s  | 1.05                    |
| lanczos (N=50)/nterms=2^18 | 4.32 ± 0.012 s    | 2.83 ± 0.022 s  | 1.53                    |
| lanczos (N=50)/nterms=2^20 | 10.6 s            | 5.84 s          | 1.82                    |
| time_to_load               | 0.303 ± 0.013 s   | 0.31 ± 0.011 s  | 0.979                   |

With 16 threads on my machine:

|                            | 9f209c0dbabaa5... | dirty           | 9f209c0dbabaa5.../dirty |
|:---------------------------|:-----------------:|:---------------:|:-----------------------:|
| lanczos (N=50)/nterms=2^14 | 0.453 ± 0.0082 s  | 0.565 ± 0.039 s | 0.801                   |
| lanczos (N=50)/nterms=2^16 | 1.38 ± 0.044 s    | 1.17 ± 0.027 s  | 1.18                    |
| lanczos (N=50)/nterms=2^18 | 4.37 ± 0.037 s    | 2.74 ± 0.024 s  | 1.6                     |
| lanczos (N=50)/nterms=2^20 | 10.5 s            | 5.52 s          | 1.89                    |
| time_to_load               | 0.293 ± 0.015 s   | 0.299 ± 0.013 s | 0.981                   |

From this I'd say that this approach does work, and seems to become more effective at larger nterms, but the actual maximal speed is bound by the dictionary insertions anyways, and it's hard to avoid the final merge dominating the cost.