Use IndexMap for performance boost in qubit allocation

[swernli]

This change pulls in the IndexMap struct from https://github.com/microsoft/qsharp/tree/main/compiler/qsc_data_structures to use for the qubit id mapping, rather than FxHashMap. In most use cases qubit ids tend to be dense and sequential, making IndexMap an optimal fit. This also cuts down on the execution time of the test_large_state unit test by an order of magnitude, and the new benchmark shows significant gains as well. Other benchmaks showed a small decrease in the perf (around 5%) but since those are measured in microseconds it seems to be a worthwile trade-off (for example, the X gate benchmark was slower by about 0.15 microsecods).

[swernli]

Flamegraphs from the unit test before the change:

after the change:

New benchmark perf:

Allocate-Release 10k qubits
                        time:   [180.63 ms 181.10 ms 181.60 ms]
                        change: [-85.182% -85.143% -85.104%] (p = 0.00 < 0.05)
                        Performance has improved.