Validation layer catch(...) behavior is confusing

I recently encountered behavior in SteamVR where [`xrAttachSessionActionSets` would sometimes throw C++ exceptions](https://steamcommunity.com/app/250820/discussions/8/3059617763557427211/). While this may be a bug in SteamVR (throwing exceptions through a C ABI is obviously not well-defined), my investigation was slowed by the validation layers' silent, blanket conversion of all caught exceptions to `XR_ERROR_VALIDATION_FAILURE`, giving the appearance that specific application misbehavior had been detected, although no diagnostics were produced. This was further confounded by (I presume) C++ ABI details causing the exceptions to only be caught in release builds of the layers.

Ideally, the validation layer should only catch its own exceptions, by e.g. inheriting them from a private base class, and then `catch`ing that base class by reference.

Start by tracing the validation layer's exception handling around xrAttachSessionActionSets and inspect how caught exceptions become XR_ERROR_VALIDATION_FAILURE. Determine how layer-owned exceptions are identified and how external C++ exceptions currently reach the catch block. Done means external exceptions are not silently reported as validation failures and the layer's own failures retain their intended result.

Written by the indexing model from the issue text.