fix multi gpu setup

[humanx98]

This PR fixes two issues that prevent HIPRT from working correctly on multi-GPU setups:

1. An Orochi bug. A minimal fix is included here to make the root cause obvious, but the preferred upstream fix is the one from ioroDevice issue on multigpu system Orochi#135, which removes bit-field usage.
2. An issue in the rtip31Support check. The Compiler constructor is executed before the device is assigned, so the support check happens too early.

[Copilot]

Pull request overview

This PR addresses multi-GPU correctness issues by (1) fixing device encoding in the vendored Orochi layer and (2) deferring the RTIP 3.1 support probe until after a context/device is set, so the check runs against the intended GPU.

Changes:

• Set the created Orochi context as current during hiprt::Context construction and explicitly initialize the compiler afterwards.
• Replace Compiler’s constructor-time RTIP support probe with an explicit Compiler::init() call performed by Context.
• Fix Orochi’s raw-device wrapping to avoid losing the device index when encoding API+device.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
hiprt/impl/Context.cpp	Sets current context during construction and triggers compiler initialization.
hiprt/impl/Compiler.h	Introduces init() and removes the explicit constructor declaration.
hiprt/impl/Compiler.cpp	Moves constructor logic into Compiler::init().
contrib/Orochi/Orochi/Orochi.cpp	Fixes raw device packing by explicitly setting API and device fields.

---

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

[meistdan]

Thank you. It looks good.