Conflicting package versions break docker build

[aguschin]

This should happen for any mlem build docker, but I got this for flyio deploy:

#7 19.23 INFO: pip is looking at multiple versions of numpy to determine which version is compatible with other requirements. This could take a while.
#7 19.23 ERROR: Cannot install -r requirements.txt (line 3) and numpy==1.21.6 because these package versions have conflicting dependencies.
#7 19.23
#7 19.23 The conflict is caused by:
#7 19.23     The user requested numpy==1.21.6
#7 19.23     tensorflow 2.6.4 depends on numpy~=1.19.2
#7 19.23
#7 19.23 To fix this you could try to:
#7 19.23 1. loosen the range of package versions you've specified
#7 19.23 2. remove package versions to allow pip attempt to solve the dependency conflict
#7 19.23
#7 19.23 ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/user_guide/#fixing-conflicting-dependencies
#7 19.35 WARNING: You are using pip version 21.1.3; however, version 23.0.1 is available.
#7 19.35 You should consider upgrading via the '/usr/local/bin/python -m pip install --upgrade pip' command.
#7 ERROR: executor failed running [/bin/sh -c pip install -r requirements.txt && pip cache purge]: exit code: 1
------
 > [4/6] RUN pip install -r requirements.txt && pip cache purge:
#7 19.23     The user requested numpy==1.21.6
#7 19.23     tensorflow 2.6.4 depends on numpy~=1.19.2
#7 19.23
#7 19.23 To fix this you could try to:
#7 19.23 1. loosen the range of package versions you've specified
#7 19.23 2. remove package versions to allow pip attempt to solve the dependency conflict
#7 19.23
#7 19.23 ERROR: ResolutionImpossible: for help visit https://pip.pypa.io/en/latest/user_guide/#fixing-conflicting-dependencies
#7 19.35 WARNING: You are using pip version 21.1.3; however, version 23.0.1 is available.
#7 19.35 You should consider upgrading via the '/usr/local/bin/python -m pip install --upgrade pip' command.
------
Error failed to fetch an image or build from source: error building: executor failed running [/bin/sh -c pip install -r requirements.txt && pip cache purge]: exit code: 1

Which is because:

requirements:
- module: numpy
  version: 1.21.6
- module: keras
  version: 2.6.0
- module: tensorflow
  version: 2.6.4

Suggestion: fall back to installing incompatible versions, since in this case it should be expected and desired. (Question: how to allow installing exact packages/versions from .mlem, but keep version resolution mechanism in place for other packages/versions?)

model to reproduce with
price_keras_preprocess_broken.zip