[Bug] .watch() can hang when RPC block num is stale

### Component

provider, pubsub

### What version of Alloy are you on?

1.7.3

### Operating System

macOS (Apple Silicon)

### Describe the bug

I've noticed that watch() on an http provider can hang for long times || indefinitely when the underlying RPC nodes have stale block nums (i.e. returned by `eth_blockNumber`).

This isn't that uncommon with node providers like Alchemy due to the way they seem to do routing.

The issue is that the way `watch_pending_transactions()` works with an http provider seems to be to get the latest block number, then iterate (via `NewBlocks`) through recent blocks, checking for existing of the tx being watched in each. As polling/processing time gets close to block time, this can hang for a long time, sometimes even indefinitely!

I've moved away from using `watch()` in favor of just locally looping `eth_getTransactionReciept`, but it might be worth doing the dumb thing of that and/or documenting on `watch()` that this inconsistency exists.