Fix MPFuture failing outside inference mode

[borzunov]

Fixes this error discovered while working on bigscience-workshop/petals#91:

[codecov]

Codecov Report

Merging #521 (1f49d50) into master (94c985d) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #521      +/-   ##
==========================================
+ Coverage   75.95%   75.97%   +0.02%     
==========================================
  Files          81       81              
  Lines        7946     7947       +1     
==========================================
+ Hits         6035     6038       +3     
+ Misses       1911     1909       -2     

Impacted Files	Coverage Δ
hivemind/utils/mpfuture.py	89.33% <100.00%> (+0.04%)	⬆️
hivemind/dht/protocol.py	93.15% <0.00%> (+0.91%)	⬆️

[borzunov]

Technically, this happens when MPFuture and the underlying shared state tensor are created inside inference_mode(), then the state is changed outside of it:

In [1]: import torch

In [2]: with torch.inference_mode():
   ...:     t = torch.zeros(3)
   ...: 

In [3]: t[...] = torch.ones(3)
---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Input In [3], in <cell line: 1>()
----> 1 t[...] = torch.ones(3)

RuntimeError: Inplace update to inference tensor outside InferenceMode is not allowed.You can make a clone to get a normal tensor before doing inplace update.See https://github.com/pytorch/rfcs/pull/17 for more details.