Support drmemtraces derived from hardware execution view

Certain tracing techniques output the hardware execution view, e.g., Intel-PT, ETM, and QEMU-TCG. Such hardware execution view is different than the regular drmemtraces in various ways:
- it is core-sharded naturally, so each trace corresponds to one hardware vcpu, unlike the regular drmemtraces which are thread-sharded.
- it may have kernel execution included by default, so some markers like kernel_xfer (that denotes a control altering syscall) make less sense since we'll have the full syscall trace too.
- it may not have the same events that we generally support in drmemtraces, like kernel_event and kernel_xfer for user signal handlers, or the various function markers.

To allow conversion to the drmemtrace format, we need to add some new notions. Particularly:
- a new offline_type_t to denote such hardware vcpu centric traces
- a new pair of markers similar to kernel_event and kernel_xfer, to denote similar events on the vcpu, such as hardware interrupts (some of these may lead to a user signal handler invoked eventually) and hardware context return (which may change exception/ring level; such as eret and iret).
- adaptations to the invariant checker to relax existing checks and perform some additional ones
- may be more...