update TensorFlow easyblock to put Bazel build files in build directory + avoid unnecessary runtime patching

[Flamefire]

Currently we used /tmp (or better: tempdir as set by EB options) to put bazel build files in. However often users set builddir to some fast storage (e.g. /dev/shm) so using that instead can speed up the build of TensorFlow. This PR solves that and also unifies some settings of various directories used

Additionally it removes some patches:

• The configure.py patch doesn't apply since 1.12.0: tensorflow/tensorflow@56a99d6 As this hasn't caused any issues yet it can be removed. It tried building some TF 2.x and 1.15 versions and no .cache/bazel folder is created
• Also I was told that TF is much more friendly to new toolchains. I hence tested removing our various, extensive patching of CROSSTOOL files with some TF 2.x ECs and it seems to be indeed no longer required. Bazel/TF seem to detect the built-in directories just well, so no need to manually specify them. I don't know when that happened (and am unsure how to find that) so I simply used 2.0 as the cutoff point. We've got 1.13.1 for 2019a, so I could test and use that as well, if wanted. Factoring this out into a separate function greatly simplifies the code making it more readable.

[Flamefire]

Test:

x86

• TensorFlow-1.13.1-foss-2019a-Python-3.7.2.eb
• TensorFlow-1.13.1-fosscuda-2019a-Python-3.7.2.eb
• TensorFlow-1.14.0-foss-2019a-Python-3.7.2.eb
• TensorFlow-1.14.0-fosscuda-2019a-Python-3.7.2.eb
• TensorFlow-1.15.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-1.15.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-1.15.2-foss-2019b-Python-3.7.4.eb
• TensorFlow-1.15.2-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.0.0-foss-2019a-Python-3.7.2.eb
• TensorFlow-2.0.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.0.1-foss-2019a-Python-3.7.2.eb
• TensorFlow-2.0.1-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.1.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-2.1.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.2.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-2.2.0-fosscuda-2019b-Python-3.7.4.eb

POWER

• TensorFlow-1.15.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-1.15.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-1.15.2-foss-2019b-Python-3.7.4.eb
• TensorFlow-1.15.2-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.0.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.0.1-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.1.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-2.1.0-fosscuda-2019b-Python-3.7.4.eb
• TensorFlow-2.2.0-foss-2019b-Python-3.7.4.eb
• TensorFlow-2.2.0-fosscuda-2019b-Python-3.7.4.eb

[Flamefire]

@boegel All tests finished successfully. The 1.13.1 fails due to easybuilders/easybuild-easyconfigs#11253

So IMO good to go.

[boegel]

I've also tested this with a wide range of TensorFlow easyconfigs, no problems encountered whatsoever, so good to go.

Thanks a lot for your efforts on this @Flamefire!