[REQ] Support launching a kernel from another kernel in Warp for dynamic task execution

[zyxia]

Description

I would like to request a feature that allows launching a Warp kernel from within another kernel.
Currently, it seems that Warp does not natively support kernel launches inside another kernel, similar to CUDA's dynamic parallelism (cudaLaunchKernel inside a device function).

The main motivation for this request is the need to handle a dynamically varying number of tasks.
In certain applications, the number of tasks is not known at launch time and needs to be determined dynamically during execution. Supporting kernel launches within a kernel would provide more flexibility in handling such cases efficiently.

Context