Allow to omit type hint in GroupBy.transform, filter, apply

[HyukjinKwon]

This PR proposes to allow to omit type hint.

1. If type hint is not given, Koalas collects data as pandas DataFrame.
   1.1. If it returns less than 1000, it returns ks.DataFrame(pdf).
   1.2. if it returns more than 1000, it just infers schema from pdf.
2. If type hint is found, uses that as return type directly.

• When the schema is inferred without type hint, index is cleanly kept.

• When the schema is given via type hint, index is lost - this is because pandas sometimes keeps the index and sometimes not. Koalas does not know without executing the given func once.

  Please see the example below:

  pdf = pd.DataFrame({"timestamp":[0.0, 0.5, 1.0, 0.0, 0.5], "car_id": ['A','A','A','B','B']})
  print(pdf.groupby('car_id').apply(lambda _: pd.DataFrame({"column": [0.0]})))

            column
  car_id
  A      0     0.0
  B      0     0.0
  

  pdf = pd.DataFrame({"timestamp":[0.0, 0.5, 1.0, 0.0, 0.5], "car_id": ['A','A','A','B','B']})
  print(pdf.groupby('car_id').apply(lambda x: x))

     timestamp car_id
  0        0.0      A
  1        0.5      A
  2        1.0      A
  3        0.0      B
  4        0.5      B
  

  Therefore, index information is lost in this case

Resolves #628
This PR also resolves 2. and 3. at #409 (comment)

[HyukjinKwon]

@patryk-oleniuk, with this PR and #633 PR, you will be able to proceed further. Hope this unblocks you.

[HyukjinKwon]

Mainly only deduplication here.

[softagram-bot]

Softagram Impact Report for pull/646 (head commit: 78c5ccc)

⭐ Change Overview

(Open in Softagram Desktop for full details)

📄 Full report

• Permalink: Full report for pull/646

Give feedback on this report to [email protected]

[codecov-io]

Codecov Report

Merging #646 into master will increase coverage by 0.29%.
The diff coverage is 62.5%.

@@            Coverage Diff             @@
##           master     #646      +/-   ##
==========================================
+ Coverage   92.97%   93.27%   +0.29%     
==========================================
  Files          31       31              
  Lines        5085     5082       -3     
==========================================
+ Hits         4728     4740      +12     
+ Misses        357      342      -15

Impacted Files	Coverage Δ
databricks/koalas/groupby.py	85.75% <62.5%> (+4.19%)	⬆️

---

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 51c0299...78c5ccc. Read the comment docs.

[HyukjinKwon]

tests passed. codecov is due to missing coverage hits in Python worker side

[HyukjinKwon]

I think I should have a configuration for such limit in a separate PR. I'll do it after this and your PR are merged @ueshin

[ueshin]

I merged my PR for the basic configuration.