[Datapath] Add initial delay optimisation pass

[cowardsa]

The datapath-reduce-delay pass applies transformations that aim to reduce the delay of any matching path in the design. The initial PR includes the following transformations:

1. Unsharing: Folding additions into datapath.compress operators even if the adder result is used by other operations - potentially increasing circuit area.
2. Mux Duplication: (sel ? a+b : c+d) + e -> (sel ? a : c) + (sel ? b : d) + e

These optimisations are enabled when targeting timing driven optimisation using circt-synth. These transformations are inspired by the work of Zimmermann on datapath synthesis (see presentation: https://picture.iczhiku.com/resource/eetop/wyiEELySpYPudnxV.pdf)

[uenoku]

LGTM, thank you for adding documents and examples as well. Looks really clever.