Spawn a second spawner for parAlgos

[hugosenari]

I was thinking about optimizations for #35 using the following logic:

Serial is T = op * N or O(nop)

proc serial(list, op):
  for i in list:
    op(i)

Parallel is T = op + schedule * N or O(1op + nspawn) ^{limited by available threads}

proc parallel(list, op):
  for i in list:
    m.spawn op(i)

• When our schedule cost is near to 0_op: O(1op).
  ie. op takes 1s, spawn takes 1us is (to be precise is O(1op + 0.000.0001*nspawn)).
• When our schedule cost is half op time: O(1op + 2/nspawn),
  ie. op takes 2us, spawn takes 1us
• When our schedule cost is equal to op time: O(1op + nspawn),
  ie. op takes 1us, spawn takes 1us

But, algorithms like README example (depth-first search) are more like O(1op + (n log nspawn))

Maybe parAlgos could have similar approach of DFS:

proc recursiveParallel(list, bulksize, op):
  if list.len == 0: return
  let (head, tail) = list.take(bulksize)
  m.spawn serial(head, op)
  for part in tail.take(tail.len / 2)
    m.spawn recursiveParallel(part, bulksize, op)

Or because previous algorithm will make pool busy of scheduling instead of busy of working, two scheduling threads is enough.

• O(1op + 1spawn + 2/nspawn)

proc parAlgo(list, bulksize, op):
  for slice in list.split(bulksize):
    m.spawn serial(slice, op)

proc parParAlgo(list, bulksize, op):
  let (l,r) = tail.take(tail.len / 2)
  m.spawn parAlgo(l, bulkSize, op)
  m.spawn parAlgo(r, bulkSize, op) # ¹

¹ _{second call to parAlgo should be made without m.spawn, but is there for readability.}

[Araq]

Nothing actionable here, closing.