consensus/beacon: simplify header split

[rjl493456442]

This PR simplifies the logic for splitting headers.

Before we verify the header batch in VerifyHeaders function, we should strictly
split the provided headers into two parts: one before the ttd(including the terminal
block) and one after the ttd.

Then we can hold the truth that the tds of pre-headers(except potential terminal
block) don't exceed the ttd, no matter the header difficulty is correct or not
(can be fake, but will be detected later in verification). And also td tds of post-headers
must all exceed the ttd(since the assumed difficulty is zero).

Then we can deliver the separated headers into corresponding verification functions
which will detect all invalid headers, e.g. difficulty is invalid.

[holiman]

LGTM