Handle attempts to get block headers at invalid heights

[yeastplume]

Twofold 'belt and suspender' fix for #3681:

• First, add an error check into get_header_hash_by_height in txhashset to return an error if the header height requested is too large for pmmr::insertion_to_pmmr_index to handle. Note there's no error handling in the core pmmr functions (which all assume arguments are correct, rightly or wrongly), so this check is placed as close to it as possible.
• Second, manually handle negative heights from the difficulty iterator in get_server_stats code. Note the call to get_header_by_height is now needed here due to earlier performance optimizations on DifficultyIterator that removed the deserialization of proof nonces and left iterator entries unable to calculate their hash.

Either fix on its own would work, but get_header_by_height should be returning an error for invalid heights, while the call to it from get_server_stats shouldn't be attempting to provide invalid heights via an incorrect cast. (Note that casting from i64 to u64 is allowed in rust regardless of value, meaning that -59i64 to u64 results in 18446744073709551557).

[tromp]

see single in-line comment

[phyro]

I have run the initial sync that broke before and it no longer does. Apart from the minor comment from tromp, the changes look good 👍 I was not able to sync the node, but I now suspect the problem is on my end with the disk.

[Hecim1984]

🤑