[seq2seq] make it easier to run the scripts

[stas00]

This PR:

• converts scripts into independent executables - easier to run. Docs updates.
• tweaks ./finetune.py and ./distillation.py to be self-contained - no need to tweak PYTHONPATH via a shell script anymore
• fixes a few more left-overs from try: import .foo; except: import foo

Let me know if there any other tweaks to make.

@sshleifer

[sshleifer]

Does python finetune.py still work? If not we have a lot of untested bash scripts to update.

[stas00]

Everything works as before. You just don't have to use python foo.py anymore. You can still use that way or the shortcut of ./foo.py.

Also there is no more need to tweak PYTHONPATH, but again, the existing tweaks still work - it just does it twice.

[codecov]

Codecov Report

Merging #7274 into master will increase coverage by 0.09%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #7274      +/-   ##
==========================================
+ Coverage   80.32%   80.42%   +0.09%     
==========================================
  Files         174      174              
  Lines       33446    33446              
==========================================
+ Hits        26867    26900      +33     
+ Misses       6579     6546      -33     

Impacted Files	Coverage Δ
src/transformers/tokenization_fsmt.py	20.34% <0.00%> (-74.90%)	⬇️
src/transformers/modeling_tf_lxmert.py	22.14% <0.00%> (-72.41%)	⬇️
src/transformers/tokenization_pegasus.py	46.03% <0.00%> (-49.21%)	⬇️
src/transformers/modeling_tf_distilbert.py	64.52% <0.00%> (-34.77%)	⬇️
src/transformers/tokenization_albert.py	70.19% <0.00%> (-23.08%)	⬇️
src/transformers/tokenization_auto.py	82.81% <0.00%> (-9.38%)	⬇️
src/transformers/modeling_gpt2.py	79.03% <0.00%> (-7.80%)	⬇️
src/transformers/configuration_utils.py	96.00% <0.00%> (-0.67%)	⬇️
src/transformers/modeling_utils.py	86.87% <0.00%> (-0.36%)	⬇️
src/transformers/data/data_collator.py	93.18% <0.00%> (-0.36%)	⬇️
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7cbf0f7...c855ebe. Read the comment docs.