refactor: move send/upgrade out of the kernel

[Stebalien]

Make them higher-level ops like everything else. I considered combining them into a single "call op" however, not all users may want to implement actor upgrading.

I've also moved the generic from the methods to the trait itself. This resolves anorth's comment on the TestKernel by allowing additional bounds to be specified by the implementer, instead of having all the bounds encoded on the trait itself:

Note that KK, the type of the kernel to crate for the receiving actor, is ignored, and Self is passed as the type parameter for the nested call. If we could find the correct bound to specify KK for the call, we would. This restricts the ability for the TestKernel to itself be wrapped by another kernel type.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (0b4a34f) 75.56% compared to head (7566332) 75.60%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1952      +/-   ##
==========================================
+ Coverage   75.56%   75.60%   +0.04%     
==========================================
  Files         157      157              
  Lines       15494    15473      -21     
==========================================
- Hits        11708    11699       -9     
+ Misses       3786     3774      -12     

Files	Coverage Δ
fvm/src/kernel/default.rs	82.33% <100.00%> (-0.10%)	⬇️
fvm/src/kernel/filecoin.rs	19.94% <ø> (-3.91%)	⬇️
fvm/src/kernel/mod.rs	90.90% <100.00%> (+2.02%)	⬆️
fvm/src/syscalls/actor.rs	89.91% <100.00%> (ø)
fvm/src/syscalls/mod.rs	97.51% <ø> (+0.49%)	⬆️
fvm/src/syscalls/send.rs	100.00% <100.00%> (ø)

... and 4 files with indirect coverage changes

[Stebalien]

After this, the only parts remaining from #1933 are:

1. Moving SyscallHandler to an associated type. I'm not actually sure if we want to do this (doesn't seem to matter in practice).
2. Refactor the syscall layering per Syscall Refactor FIPs#845. But I'm not sure if I've gotten that layering right, so I'd rather take it one step at a time.

[fridrik01]

Only one minor comment, but otherwise lgtm.

Great to move the send/upgrade into these ops to make things more flexible (and remove boilerplate code)