Support complex dtypes

[Randl]

Following discussion in patrick-kidger/diffrax#326 (comment)
From the first glance, the following is the minimum list of things to do for complex number support:

• Add dtype parameter to tests, test with both real and complex number (Add explicity dtype parameter to tests #63)
• Fix typing, specifically from Float to Inexact; add casts(Change type annotation to allow complex types #59, Add casts in additional place to satisfy strict typing for complex case #89, Norm should return float rather than complex #91)
• Add holomorphic to appropriate places in jaxfwd (Add holomorphic parameter #60)
• Add complex-related tags and methods, e.g., conj (as opposed to transpose) (Add conj function and adjoint tests #62)
• Fix complex-specific bugs in methods, e.g., missing conjugates, and make sure tests are passing (Fixes for complex-related changes (mostly conjugates) #61, Fix the inner product order in BiCG, add abs to CG #90)
• Add complex-specific and new functionality (such as adjoint) tests (Add conj function and adjoint tests #62)
• Add documentation and examples that include complex dtypes (Add complex-related stuff to docs and examples #92)
• Update the jax dependency version (related to [BUG] Inconsistent compiled JAX running results on CPU openxla/xla#8471, Update jax dependency #93)

[Randl]

I'm closing this now as all the checklist is checked. Any further improvements should go to separate issues, I think.

[patrick-kidger]

Agreed!
Seroiously, great job pushing this all the way through. I'm really happy to have this in -- and I've just pulled the trigger for the next release of Lineax!