Prepare to support PAI prediction using experimental codegen

[sneaxiy]

• Add experimental.getModelMetadataFromOSS in the Go side. Also, move pai.getModelBucket to model.GetOSSModelBucket and move pai.getModelPath to model.GetOSSModelPath.
• Add table_ops.create_tmp_tables_guard in the Python side to replace try...finally clauses.

TODO: make the runtime/pai Python package to support the experimental codegen.

[codecov-io]

Codecov Report

Merging #3039 into develop will increase coverage by 0.14%. The diff coverage is 23.39%.

@@             Coverage Diff             @@
##           develop    #3039      +/-   ##
===========================================
+ Coverage    50.18%   50.33%   +0.14%     
===========================================
  Files          251      251              
  Lines        17936    18465     +529     
===========================================
+ Hits          9002     9295     +293     
- Misses        8240     8465     +225     
- Partials       694      705      +11     

Impacted Files	Coverage Δ
go/codegen/experimental/codegen_step.go	22.45% <0.00%> (-0.25%)	:arrow_down:
go/executor/alisa.go	2.65% <0.00%> (+0.34%)	:arrow_up:
go/executor/pai.go	2.31% <0.00%> (+0.08%)	:arrow_up:
go/executor/pai_local.go	0.00% <0.00%> (ø)
go/model/model.go	32.92% <0.00%> (+0.84%)	:arrow_up:
python/runtime/model/modelzoo.py	24.24% <0.00%> (-6.53%)	:arrow_down:
python/runtime/pai/submitter_evaluate.py	23.91% <0.00%> (ø)
python/runtime/pai/submitter_explain.py	25.92% <0.00%> (ø)
python/runtime/step/tensorflow/train.py	21.97% <0.00%> (+0.92%)	:arrow_up:
python/runtime/pai/submitter_predict.py	30.43% <9.52%> (+1.54%)	:arrow_up:
... and 16 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 463a08d...975d051. Read the comment docs.