[otenim]

Summary

• Change intensity field type from std::uint8_t to float in all three output point types: PointXYZI, PointXYZIT, PointEnXYZIT (point_types.hpp).
• Drop the lround + uint8_t cast in SeyondPCDDecoder::point_xyz_data_parse; the clamped float value is written directly to point.intensity.
• Update README and docs/seyond_intensity_scaling.md to describe intensity as float32.

Robin W intensity scaling (× 255/4095 for major ≤ 3, × 255/255 for major ≥ 4) and the [0, 255] clamp are unchanged — only the storage type changes. Values remain in [0, 255] but now retain fractional precision.

Output schema impact

sensor_msgs/PointCloud2.fields will report intensity with datatype = FLOAT32 (7) instead of UINT8 (2). Downstream consumers that assume uint8 will need to read the field as float32.

Test plan

• colcon build --packages-select bag_converter succeeds (no new warnings/errors).
• bag_converter --help runs.
• Run conversion on a representative mcap and verify the output PointCloud2 reports intensity as FLOAT32 with values in [0, 255].
• Verify Robin W scaling output range/scale matches the previous uint8 rounding (within rounding tolerance).
• Plot intensity distribution with scripts/plot_field_distribution.py on Robin W and non-Robin W bags.