separate constants for average and worst case relay headers

[svyatonik]

The idea is to move bridge fees towards expected after paritytech/polkadot-sdk#2139 . There I've used existing AVERAGE_HEADER_SIZE_IN_JUSTIFICATION constants, which are actually worst possible header size. Once this PR is merged, we'll keep using worst possible const for computing maximal refund, but for computing fees we'll be using average header size. Side effects:

• the user will pay smaller fee for over-bridge messages. Let this fee be FEE;
• SA of AH at BHs will keep paying to relayers at most MaxFEE, which is larger than the FEE;
• if message proofs will be built using large headers, the difference MaxFEE - FEE will be paid by the SA of AH.

In real world, however, we'll be running our own realyer (at least for v1), which is configured to deliver messages/confirmations asap and will select the header without looking at its size. Since we expect headers to have the AVERAGE_HEADER_SIZE_IN_JUSTIFICATION, not the WORST_HEADER_SIZE_IN_JUSTIFICATION. So the actual fee in average will be closer to the FEE, not the MaxFEE.

[bkontur]

looks like you need to add WORST_HEADER_SIZE_IN_JUSTIFICATION to the polkadot-bulletin imports

pub use bp_polkadot_core::{
	AccountAddress, AccountId, Balance, Block, BlockNumber, Hash, Hasher, Header, Nonce, Signature,
	SignedBlock, UncheckedExtrinsic, AVERAGE_HEADER_SIZE_IN_JUSTIFICATION,
	EXTRA_STORAGE_PROOF_SIZE, MAX_HEADER_SIZE, REASONABLE_HEADERS_IN_JUSTIFICATON_ANCESTRY,
};