Witgen: Remove sequence cache for block machines

[georgwiese]

Fixes #1559

The sequence cache is often causing trouble, and I wonder how big the performance benefit even is after #1528 (and given that we don't spend that much time in secondary machines).

We should benchmark this PR and see what the impact is. If it is small, we can through away most of executor/src/witgen/sequence_iterator.rs!

[georgwiese]

On the Keccak RISC-V example, the block machines do seem to be roughly 2x slower. These are two representative runs:

========== BEFORE
 == Witgen profile (1766802 events)
   39.6% (    4.1s): FixedLookup
   37.7% (    3.9s): Main Machine
   12.7% (    1.3s): Secondary machine 0: main_binary (BlockMachine)
    6.3% ( 652.5ms): witgen (outer code)
    3.1% ( 319.1ms): Secondary machine 2: main_shift (BlockMachine)
    0.6% (  57.7ms): Secondary machine 1: main_memory (DoubleSortedWitnesses)
    0.0% ( 251.6µs): Secondary machine 3: main_split_gl (BlockMachine)
  ---------------------------
    ==> Total: 10.349325541s

========== AFTER
 == Witgen profile (3086124 events)
   35.4% (    4.6s): FixedLookup
   29.8% (    3.8s): Main Machine
   23.3% (    3.0s): Secondary machine 0: main_binary (BlockMachine)
    5.7% ( 738.2ms): Secondary machine 2: main_shift (BlockMachine)
    5.3% ( 677.5ms): witgen (outer code)
    0.4% (  52.3ms): Secondary machine 1: main_memory (DoubleSortedWitnesses)
    0.0% ( 335.5µs): Secondary machine 3: main_split_gl (BlockMachine)
  ---------------------------
    ==> Total: 12.854500542s

[georgwiese]

#1562 is a better solution.