add fuseLoops macro to fuse loops in Nim

[Vindaar]

Fusing the loops in this context means the equivalent of OpenMP's collapse statement, i.e. merging multiple nested loops into a single loop.

One can use the nofuse argument to indicate a nested loop should not be fused.

The following code shows the essentials:

fuseLoops:
  for i in 0 ..< N:
    let x = i * 2
    for j in 0 ..< T:
      let z = x * j
      echo i, j, x, z
    echo x
# Becomes:
# for idx_553648168 in 0 ..< N * T:
#   let i = idx_553648168 div T
#   let j = idx_553648168 mod T
#   let x = i * 2
#   let z = x * j
#   echo i, j, x, z
#   echo x


fuseLoops:
  for i in 0 ..< N:
    let x = i * 2
    for j in 0 ..< T:
      let z = x * j
      echo i, j, x, z
      for k in nofuse(0 ..< T):
        echo k
    echo x
# Becomes:
# for idx_553648190 in 0 ..< N * T:
#   let i = idx_553648190 div T
#   let j = idx_553648190 mod T
#   let x = i * 2
#   let z = x * j
#   echo i, j, x, z
#   for k in nofuse(0 ..< T):
#     echo k
#   echo x

fuseLoops("parallel for"):
  for i in 0 ..< N:
    let x = i * 2
    for j in 0 ..< T:
      let z = x * j
      for k in 0 ..< X:
        echo i, j, k, x, z
    echo x
# Becomes:
# for idx_553648204 in `||`(0, N * (T * X), "parallel for"):
#   let i = idx_553648204 div X * T
#   let j = idx_553648204 mod X * T div X
#   let k = idx_553648204 mod X * T mod X
#   let x = i * 2
#   let z = x * j
#   echo i, j, k, x, z
#   echo x

## The following raises a CT error
when compiles((
  fuseLoops:
    for i in 0 ..< N:
      let x = i * 2
      var zsum = 0
      for j in 0 ..< T:
        let z = x * j
        zsum += z
        echo i, x, z
      echo x
      for j in 0 ..< 2 * T:
        zsum += j
      echo zsum
)):
  doAssert false

[Clonkk]

we should export fuse_loops in root scinim :)

[Vindaar]

Merging despite CI failures. At least the CI failures are now different! 🤣