Now kernel object model is built with original Pid and device id, so multinode executions with single-reports have unique information of the device id attached to. This fixes #23 #15

Version bump 0.3.1. Also some new chekings relative to versioning of nsys used to profile, and schema version.

Resolve "Different NVTX event types for different domains"

Closes #16

See merge request !6

Now NVTX pushpop ranges are separated in domains, with each domain a different event type. This also means that NCCL regions have its own event type, enabling having custom information. A custom event type is also added for nesmik instrumentation.

This commit classifies MPI events with the same event types and event values as Extrae does. THis allows using same CFGs that paraver includes for MPI. It also adds Rank information (when available) to the process model building, making rank 0 first in the list.

Blocked writing with progress bars

See merge request beppp/heka-tooling!4

Changes all writing loops to a generalized function with lambda for serializing the final prv line. This closes #9.

Feature: translate NVTX Start-End ranges too

See merge request beppp/heka-tooling!3