Improve VDB Fillet SOP naming and improve performance on tree construction implementation

[apradhana]

This PR has targets two things

• Parameter name in VDB Fillet SOP
• Make the output construction to be more efficient
  For segment construction:

Case	Previous segment_build	Fused segment_build	Speedup
23M	0.0336s	0.0205s	1.64x
100M	0.129s	0.0532s	2.43x

The algorithm now does this:

• It first walks both lhs and rhs internal nodes together:
  • regions that do not overlap are copied directly into the output
  • regions that do overlap are added to primaryLeafNodes or secondaryLeafNodes
• Then one leaf pass processes those queued leaves:
  • primaryLeafNodes handles lhs-owned work, including fillet computation
  • secondaryLeafNodes handles rhs-only leaves