ioroDevice issue on multigpu system

[jpola-amd]

Fix UB type-punning in ioroDevice and packed device handle boundaries

• Replace the ioroDevice bitfield struct with a single oroU32 m_value representation. The old constructor wrote through an int* alias over bitfield storage, which is undefined behaviour under the strict aliasing rule and also relied on unspecified bitfield layout.

• The old oroSetRawDevice additionally misinterpreted a raw backend device index as an already-packed Orochi handle, causing incorrect results for any device index with bits overlapping the API field (e.g. oroDevice = 1). This appears when testing HIPRT with HIP_VISIBLE_DEVICES="1".

Changes:

• ioroDevice now stores a single oroU32; API and device index are packed/unpacked with explicit masks and shifts. API field widened from 4 to 5 bits to cover the full oroApi enum including ORO_API_CUDARTC = 1 << 4.
• oroSetRawDevice packs from zero, so the raw device index is written into the correct field rather than being misread as packed data.
• All oroDevice boundary sites (oroDeviceGet, oroGetRawDevice, oroGetDeviceProperties, oroDeviceGetName, oroDeviceGetAttribute, oroCtxCreate) replace *(ioroDevice*)device = d and *(oroDevice*)&d casts with plain (oroDevice)d.packed() / ioroDevice((oroU32)dev) conversions.

[Copilot]

Pull request overview

This PR fixes undefined behavior and incorrect packing/unpacking of oroDevice handles on multi-GPU systems by replacing the previous bitfield + type-punning approach with an explicit 32-bit packed representation.

Changes:

• Replaced ioroDevice bitfields with a single oroU32 m_value plus explicit masks/shifts for API + device index packing.
• Fixed oroSetRawDevice to pack a raw backend device index into the correct bit range (instead of treating it as already-packed data).
• Removed pointer reinterpret-casts at oroDevice boundaries and replaced them with value-based conversions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.