Allow passing a custom random number generator to trace.RandomIdGenerator

[agcom]

Description

I am using OpenTelemetry to trace gRPC calls in my distributed machine learning application, and I fixed the global random seed in all replicas via random.seed(42) to make the application's result reproducible; but, it causes duplicate trace and span ids rendering the traces corrupt.

This PR adds the possibility to create/use an instance of trace.RandomIdGenerator that utilizes a custom random number generator instance (e.g. random.Random()); It preserves backward compatibility and previous users expectations by defaulting to the global random instance as before.

Fixes #4376.

Type of change

• New feature (non-breaking change which adds functionality)

How Has This Been Tested?

• Ran the existing unit tests.

To reproduce my use-case, write a Python script with the following pseudo-algorithm:

• Fix the global random seed.
• Instrument the gRPC server and client (using the default id generator), exporting to an OpenTelemetry Collector instance which in-turn optionally exports them to a Jaeger instance.
• Spin up a gRPC server with a simple NOP function.
• Every x seconds, randomly choose a peer (assume running multiple instances of the script with pre-determined addresses) and call its function.
• Run multiple instances of the script.
• Find the duplicate ids in the output traces (can be done through the Jaeger UI which visibly warns about duplicate ids).

Does This PR Require a Contrib Repo Change?

• Yes.
• No.

Checklist:

• Followed the style guidelines of this project.
• Change-logs have been updated.
• Unit tests have been added.
• Documentation has been updated.

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: agcom / name: Alireza Ghasemi (9fedfbc, 7af0a73, dc5133b)
• ✅ login: lzchen / name: Leighton Chen (0bc5a8b)

[agcom]

Ideally, the default RNG should be a fresh Random() (which picks a high-entropy seed through combining /dev/urandom, the current time, and the process identifier) and not the global random instance; but, that would be a breaking change/update, affecting the previous codes and users expectations relying on that behavior (unlikely, but yet possible).

[agcom]

I am not sure of what unit tests to add if they are mandatory for this change. In the tests/trace/test_trace.py file, there are lines where an instance of RandomIdGenerator is created and used to generate a few ids; I could change them all to RandomIdGenerator(Random()).