Increment write backpressure before deferred encryption

[mkeeter]

(Staged on top of #1443 )

This PR tightens the feedback loop for backpressure, by incrementing our counters before encryption is complete. It uses the new RAII BackpressureGuard type, so we don't need to worry about weird accounting failures.

Before:

                ┌────────────┐ incr             
             ┌──┤backpressure│◄───┐             
             │  └────────────┘    │             
┌─────┐  ┌───▼──┐    ┌──────────┐ │    ┌───────┐
│Guest├─►│submit├───►│encryption├─┼───►│clients│
└───▲─┘  └──────┘    └──────────┘ │    └───────┘
    │                             │             
    └─────────────────────────────┘             
              ack                               

After:

                ┌────────────┐                  
             ┌──┤backpressure│                  
             │  └─▲──────────┘                  
             │    │incr                         
┌─────┐  ┌───▼──┐ │  ┌──────────┐      ┌───────┐
│Guest├─►│submit├─┴─►│encryption├─┬───►│clients│
└───▲─┘  └──────┘    └──────────┘ │    └───────┘
    │                             │             
    └─────────────────────────────┘             
              ack                               

[mkeeter]

This failed CI in test-up-encrypted: https://github.com/oxidecomputer/crucible/runs/29621235532

I'm rerunning now to see if it was a flake or reproducible