fix: Allow to display events on devices

[speeddragon]

Since the restructuring of devices (#915), setting a device name in HB_PRINT stopped working. Eg HB_PRINT=dev_manifest.

This PR gets the device name from HB_PRINT and finds the device name that is used inside the BEAM.
The format is device@version. Eg manifest@1.0.

[samcamwilliams]

Why must the agents try to solve every problem in Erlang by the introduction of yet-another-adhoc persistent_term cache? 😂

Can you see any other ways to solve this? We could at minimum at least do this in hb_device's load flow where we have the existing name@ver -> module caching?

[speeddragon]

Even before asking the agent, this solution came to mind. Other options would be ETS or adding it to Opts. ETS is 20% slow than persistant_term and modifying Opts doesn't feel like a good solution.

Agree, the persistant_term should be done on module load and not on an event call, which can be subject to concurrency.

[samcamwilliams]

ets vs persistent_term point is interesting. Should we benchmark persistent_term against ets for hb_store_volatile? We use it a ton.

[speeddragon]

I don't think persistent_term is a good candidate where too much writing is needed, because it triggers underlying code on each write. I asked an agent to run a benchmark.

On this machine, tuned ETS achieved roughly 13.6M parallel writes/s versus 387K for persistent_term. persistent_term remained competitive for parallel reads.

[speeddragon]

Updated the PR to follow device@version format and change only hb_opts file.
Also in this PR there is a fix for calling default_message_with_env and default_message.
Changing erlang:get to persistent_term:get, since erlang:get only applies to a single process, not the whole VM. This means this was calling for every request.