cuprated: Add graceful shutdown

[redsh4de]

What

Ports over the graceful shutdown logic from fjall-wip branch: https://github.com/Cuprate/cuprate/blob/fjall-wip/binaries/cuprated/src/monitor.rs

Depends on the embeddable version #592

Why

More idiomatic shutdown pattern, previously every exit was equivalent to a crash

Where

• cuprated:
  • monitor.rs - TaskExecutor handle (cancellation token + task tracker)
  • config.rs - target_max_memory / fjall_cache_size / block_downloader_config return Result instead of panicking on unresolved memory
  • main.rs - wires up executor, OS signal handler, double-Ctrl+C force exit, shutdown wait loop
  • commands.rs - add Exit command
  • blockchain/manager.rs, syncer.rs - shutdown token in select loops, replace todo!() with break
  • txpool/manager.rs, incoming_tx.rs - shutdown token, change ordering by priority due to using biased
  • rpc/server.rs - with_graceful_shutdown on axum, tracked spawns, shutdown token plumbed through
  • rpc/handlers/other_json.rs - wiring for stop_daemon RPC
  • lib.rs - Node::launch returns Result, creates and distributes TaskExecutor, Drop impl
• storage/blockchain:
  • database.rs - detach drop-log info! from caller span

How

monitor.rs introduces TaskExecutor, took nomenclature inspo from reth. Instead of how reth does it though with a static, its threaded into the functions to support running multiple Node instances in the same process

All tokio::spawn replaced with TaskExecutor::spawn. Task select loops now use biased ordering with the shutdown token as higher priority.

On shutdown signal, the token is cancelled -> tasks wind down one by one, and main() waits on task tracker to finish unwinding before dropping the runtime.

Node::launch now returns Result<Self, anyhow::Error> instead of panicking on database/service init failures.

Double Ctrl+C force exits.

[redsh4de]

Reworking this to depend on #592, ignore for now

[redsh4de]

Depends on #592 now.

[SyntheticBird45]

I let a few comment but overall lgtm