Improve block_sim speed with 50k+ validators

[tersec]

More progress towards running 50k or 100k validators smoothly on block_sim.

• get_beacon_committee() was pointlessly calling the full version of get_committee_count_at_slot(), when it's just a simple arithmetic function of the number of active validators, which get_beacon_committee() already has.

• get_inclusion_delay_deltas() is part of the epoch processing, which had a quadratically growing part.

• the get_committee_assignment() / installValidatorApiHandlers() changes prepare for dynamically subscribing and unsubscribing to attestation subnets.

Before:

$ time build/block_sim --validators=100000 --slots=64
Loaded genesim_mainnet_100000_0.12.1.ssz...
Starting simulation...
................................ slot: 32 epoch: 1
................................ slot: 64 epoch: 2
Done!
Validators: 100000, epoch length: 32
Validators per attestation (mean): 0.0
All time are ms
     Average,       StdDev,          Min,          Max,      Samples,         Test
    4025.887,     2255.056,     2237.409,    15992.623,           62, Process non-epoch slot with block
   21661.675,    16770.872,     9802.878,    33520.473,            2, Process epoch slot with block
       4.948,        0.878,        0.018,        5.684,           64, Tree-hash block
       1.562,        0.103,        1.385,        1.842,           64, Sign block
    7754.820,      156.415,     7471.934,     8206.023,           64, Have committee attest to block
       3.745,        0.000,        3.745,        3.745,            1, Replay all produced blocks

real	13m29.612s
user	13m18.196s
sys	0m2.752s

After:

$ time build/block_sim --validators=100000 --slots=64
Loaded genesim_mainnet_100000_0.12.1.ssz...
Starting simulation...
................................ slot: 32 epoch: 1
................................ slot: 64 epoch: 2
Done!
Validators: 100000, epoch length: 32
Validators per attestation (mean): 0.0
All time are ms
     Average,       StdDev,          Min,          Max,      Samples,         Test
    3859.758,     2263.392,     2142.557,    16094.623,           62, Process non-epoch slot with block
   11636.219,     3293.731,     9307.200,    13965.239,            2, Process epoch slot with block
       4.811,        0.852,        0.023,        5.823,           64, Tree-hash block
       1.521,        0.087,        1.355,        1.764,           64, Sign block
    4311.929,      142.398,     3978.022,     4696.124,           64, Have committee attest to block
       3.924,        0.000,        3.924,        3.924,            1, Replay all produced blocks

real	9m16.943s
user	9m9.943s
sys	0m1.356s

Non-epoch slot times only go down slightly -- that was worth fixing, but was coming up when reloading an epochRef into StateCache, where the latter was being incompletely repopulated due to EpochRef not containing all the same information. In particular, it was coming up in onAttestation, which doesn't have any other source for its cache.

Epoch slot times are reduced by 2 to 3 times.

The combination creates a ~40% increase in block sim speed overall at 100k validators. It doesn't seem to matter nearly as much up to, say, 20k validators.