feat: initial IFRT integration

[avik-pal]

I will split up this PR into multiple smaller PRs once this is ready

• ReactantExtra changes
• Move certain functions into XLA.PJRT
• Fixes to the global_ordinals
• Distributed Runtime feat: setup Distributed Runtime #780
• Sharding C-API feat: use C-API to convert TensorShardingAttr to HloSharding #788

[mofeing]

changes are being made to the C-API, and specifically on the way we should make interact PjRt and IFRT (due to discussions in #738, where I was adding the Julia API). see #751 and specifically this comment #751 (comment)

we are overlapping a bit: i was waiting on #751 to push some of the API you're adding here 😅. i think we should first fix this #751 (comment) and then add the Julia API. i have all the API you want locally but we need first to fix this comment i left, as otherwise sharding won't work for distributed TPU (which is why we want IFRT in the first place).

EDIT: just remembered we agreed with @wsmoses to start trying just with IFRT-PjRt backend for testing, but we must be careful as this PR conflicts with #751 (where i'm changing the C-API) and with #738 (where I introduce the Julia bindings which is blocked on #751).

[avik-pal]

I am not adding any PjRt-IfRt interaction stuff (Held* functions in that PR) here (except the ifrt::Client constructed from the PjRtClient).

We need to use the direct IFRT calls for constructing arrays from HloSharding which doesn't take PjRtBuffers (at least not without multiple roundtrips)

[avik-pal]

I0000 00:00:1740173243.405339 1677674 pjrt_client.cc:524] PjRt-IFRT device count: total=4, addressable=2
I0000 00:00:1740173243.405393 1677674 pjrt_client.cc:528] Addressable PjRt-IFRT device: CpuDevice(id=0)
I0000 00:00:1740173243.405399 1677674 pjrt_client.cc:528] Addressable PjRt-IFRT device: CpuDevice(id=1)
I0000 00:00:1740173243.405382 1677675 pjrt_client.cc:524] PjRt-IFRT device count: total=4, addressable=2
I0000 00:00:1740173243.405436 1677675 pjrt_client.cc:528] Addressable PjRt-IFRT device: CpuDevice(id=131072)
I0000 00:00:1740173243.405442 1677675 pjrt_client.cc:528] Addressable PjRt-IFRT device: CpuDevice(id=131073)
Reactant.XLA.process_index(client) = 1
Reactant.XLA.process_index(client) = 0
1 CPU:0 cpu
0 CPU:0 cpu
1 CPU:1 cpu
1 CPU:131072 cpu
1 CPU:131073 cpu
0 CPU:1 cpu
0 CPU:131072 cpu
0 CPU:131073 cpu
1 CPU:131072 cpu
1 CPU:131073 cpu
0 CPU:0 cpu
0 CPU:1 cpu

CPU setup for distributed (and presumably TPU setup) now works