Controllable precision when printing

[yuanming-hu]

Concisely describe the proposed feature

import taichi as ti

ti.init()

@ti.kernel
def foo():
    a = 1.23456
    print(f"{a:.2f}")

foo()

The user would expect 1.23.

Describe the solution you'd like (if any)
IIRC Taichi stores the numbers in a buffer and format them on the host. Perhaps we should allow a richer syntax?

Also note that we need support for vector/matrix and even structs.

Additional comments
Not sure if this feature request was already raised somewhere...

[ailzhang]

FYI this feature can be added without too much hassel since it's supported on both llvm and spirv based backends.
reference:

• https://github.com/KhronosGroup/Vulkan-ValidationLayers/blob/master/docs/debug_printf.md
• https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html
  This issue can be a pretty good get started issue for a new contributor who are interested in taichi codegen part. cc @gingerkidney

Adding this to v1.2.0 milestone so that we can get back to it in the future if no one picks it up before that.

[ailzhang]

cc: @dream189free