feat: params.ChainConfig extra payload can use root JSON

[ARR4N]

Why this should be merged

Adds support for backwards compatibility with JSON format of ava-labs/{subnet-evm,coreth}/params.ChainConfig.

How this works

The params.Extras.ReuseJSONRoot bool flag is added to allow the end user to signal the need for backwards compatibility. This isn't the default because of the need for multi-pass (un)marshalling.

For extra payload:

type ChainConfigExtra struct {
  Foo string `json:"foo"`
}
//...
params.RegisterExtras(Extras[ChainConfigExtra,...]{
  ReuseJSONRoot: ...
})

ReuseJSONRoot == false =>

{
  "chainId": 42,
  "extra": {"foo": "bar"}
}

ReuseJSONRoot == true =>

{
  "chainId": 42,
  "foo": "bar"
}

How this was tested

Unit tests for JSON (un)marshalling for all cases; i.e when:

1. No registered extras: plain JSON (un)marshalling;
2. Flag not set: as before this PR, "promoting" the extra field to be exported with the json: "extra" tag; and
3. Flag set: feature added in this PR.

[darioush]

I noticed in a couple cases that MarshalJSON() being defined on the value vs. pointer makes a difference.
Since it is only a couple instances so far I switched the uses to & to move on, but perhaps this should be defined on the value to avoid invoking the default MarshalJSON when the receiver is non-pointer.

[ARR4N]

Interesting! Do you remember where this was, so I can take a look before I merge this? I defined it on the pointer because that's what tends to be carried around, e.g. in vm.EVM.

[ARR4N]

I'm going to leave it as defined on the pointer for now. In the geth code base there are only two instances where a non-pointer ChainConfig is used (note that most of the results are either new(ChainConfig) or immediately address the value with &).

This is in contrast to the 65 files with a *ChainConfig.

[darioush]

The *ChainConfig instances will still work if the receiver is value. We can leave it as is for now.

[ARR4N]

Sorry about pushing the refactor in a9250c6 after your review; I didn't realise that you'd already approved. The code is functionally identical and much simpler, plus there are no changes to the tests so I'll continue to merge it.