fix: declare the core.lock to be a pointer to make core concurrency safe

[ZetaoZhuang]

Reason for Change:

To make the AI Core concurrency safe, we need to declare the core.lock to be a pointer so in cloned core we can point to the same lock
Issue Fixed:

Requirements:

• uses conventional commit messages
• includes documentation
• adds unit tests

Notes:

[rbtr]

@timraymond that this is necessary feels like something is big wrong, wdyt?

[timraymond]

@rbtr well, the patch is correct as it stands in isolation, because bd236e7 introduced a defect to zapai.(*Core).clone() where the mutex is copied. I guess what I don't understand is why the mutex exists in the first place. It's not clear what it's protecting...

[ZetaoZhuang]

@timraymond currently we are only able to have a single encoder shared between a core and its derived. so the mutex here is to protect the traceTelemetry of encoder while there is a process working on resetting/updating it

[ZetaoZhuang]

@timraymond currently we are only able to have a single encoder shared between a core and its derived. so the mutex here is to protect the traceTelemetry of encoder while there is a process working on resetting/updating it

@rbtr any thoughts on this?

[rbtr]

@ZetaoZhuang I regret the singleton gobber, I don't think this is sustainable. I discussed this with @timraymond who had the idea that maybe we could use protobuf to encode/decode the TraceTelemetry<->[]byte which could get us away from this synchronization mess...wdyt? Could you investigate some alternatives?

[timraymond]

@rbtr btw, I approved this because the code in its current state is provably broken, and this patch does provide a minimal fix. I agree that we should investigate using protobufs, but I don't think this PR should be a forcing function for that investigation. I'll open an issue with the idea so that it can be handled separately.

[ZetaoZhuang]

@ZetaoZhuang I regret the singleton gobber, I don't think this is sustainable. I discussed this with @timraymond who had the idea that maybe we could use protobuf to encode/decode the TraceTelemetry<->[]byte which could get us away from this synchronization mess...wdyt? Could you investigate some alternatives?

+1 on the protobuf. we can do the investigation on protobufs separately.

[rbtr]

no objection