Insert Algorithm

Two phases:

1. Walk to find the insertion point at each level — same as search, but record the predecessor at each level in update[].
2. Splice the new node in by swapping forward pointers, level by level.

The random_level() function flips coins to decide how tall the new node is. With p=0.5: ~50% are level 0, ~25% level 1, ~12% level 2, etc. Expected average pointers per node: 2.

If duplicate keys are allowed: append (with same key, after existing matches). For unique-key set, replace value if found.