handle the HF invalid header scenario safely during chain init

[antiochp]

Replaces #2910.

[tbd - description of why this necessary with some context around the floonet HF]

This PR -

• txhashset extension and header_extension how take a head (chain tip) and not a header
• build the head MMR in parallel with the sync MMR (if we increase total work)
• refactor, pull out setup_genesis_head
• cleanup and simplify update_head and update_header_head
• make setup_head more robust in terms of rewinding if we cannot read header from db
• input_pos_to_rewind (to rewind "spent" UTXO pos) now takes a height and can operate over a set of input_pos without needing to read headers from the db
• reader header hashes from the header (and sync) MMR so we can determine which headers we need to rewind without necessarily needing to read/deserialize them from the db

TODO -

• clean up error! logging

---

This got too big and unwieldy again.
Related - #2918 (which should go a long way toward making this fix simpler to impl).

[antiochp]

Closing - will replace with smaller PRs.