server/assets/eth: refactor options for rpclient bestHeader

[buck54321]

There are 2 options here

The (*rpcclient).bestHeader method was not optimized to reduce rpc calls. I've given two options to make it better.

1. We don't need provide a block number to HeaderByNumber. The arg is a *big.Int and if we pass nil, it is parsed to "latest" which has a special meaning for geth. This is the first commit.
2. The load of getting a header is large. We could still call BlockNumber, but only call HeaderByNumber if the tip height has changed. This is the second commit.

[chappjc]

With #2105, I think we're good without the cache in the second commit.

I think your use of the nil arg in #2104 is good.

Seems like we already have all the bases covered in those.

[chappjc]

Still, I don't see a problem with the cache commit. It could make the difference when RPCs really need to be light 🤷

[buck54321]

bestHeader is called once per market on startup via (Swapper).ChainsSynced -> (*baseBackend).Synced. It's also still called in FeeRate, but it's not clear to my why we even have that method on eth, since it's not really used by the feeFetcher. I guess just cause it's part of the Backend interface.

Anyway, no cache needed, I think. Closing.