Improved Extranonce Prefix Generation

[coleFD]

The current method for extranonce_prefix generation simply increments a uint, which has size/resiliency trade-offs. Theoritically the extranonce_prefix for a single server only needs 2 bytes since the servers will not run more than 65k connections. Incrementing a u16 for extranonce prefix is not wise because connections drop and reconnect all the time meaning extranonce_prefixes will wrap on saturation and get reused which duplicates work. To fix this, the current method allocates a larger extranonce prefix so that it will never wrap, but this comes at the cost of coinbase space.

A more efficient method is to use a constant size bitvec with length equal to the number of connections the stratum server would like to allow. When a client connects, the extranonce prefix will be the index of an open slot in the bitvec. When a client disconnects, the slot can be freed so future connection can reuse the extranonce prefix.

[plebhash]

in #1944 we're being forced to introduce some checks on JDC's handle_set_extranonce_prefix

the current APIs exposed by mining_sv2::ExtendedExtranonce will not accept a range_2 that goes beyond MAX_EXTRANONCE_LEN (which is 32 bytes`)

but there's actually nothing that prevents a full_extranonce_len from being bigger than 32 bytes, as long as it doesn't violate the 100 byte limit of scriptSig together with everything else that goes in there (which is already enforced by channels_sv2 server channel constructors and JobFactory)

cross-referencing #1944 (comment)