Design a Configurable Point-Cloud Transformation Pipeline
Design an in-process library that validates a YAML-defined sequence of point-cloud transformations and executes it safely in order.
Design the low-level architecture for a point-cloud preprocessing library that reads a YAML configuration containing an ordered sequence of transformations and applies the resulting pipeline to point-cloud frames.
For this exercise, assume a configuration has a schema version, a pipeline name, and uniquely identified ordered stages; each stage names a registered transformation kind and its parameters. Configuration is parsed, validated, and compiled before any frame is processed.
A frame run must either return one completed result or one structured failure, never a caller-visible partial result. New transformation kinds and supported schema versions should be addable without rewriting pipeline orchestration.
The transformations are in-process and do not retain frame-specific mutable state between calls. YAML parser internals, distributed scheduling, persistence, point-cloud file I/O, and the numerical implementation of individual transformations are out of scope. Produce a design, not production code.