[Refactor]: improve performance for detecting function components

[golopot]

Benchmark on a private repo with time npx eslint . (real time):

	before	after
run 1	6.6s	4.8s
run 2	6.5s	4.7s
run 3	6.7s	4.8s

Benchmark with TIMING=10 npx eslint .:

Before:

Rule                                | Time (ms) | Relative
:-----------------------------------|----------:|--------:
react/destructuring-assignment      |   345.923 |     7.0%
react/no-unstable-nested-components |   208.838 |     4.2%
react/no-did-mount-set-state        |   189.097 |     3.8%
react/boolean-prop-naming           |   187.663 |     3.8%
react/no-direct-mutation-state      |   186.664 |     3.8%
react/no-deprecated                 |   184.020 |     3.7%
react/hook-use-state                |   146.379 |     3.0%
react/no-access-state-in-setstate   |   145.210 |     3.0%
react/prefer-stateless-function     |   133.564 |     2.7%
react/require-optimization          |   128.301 |     2.6%

After:

Rule                                | Time (ms) | Relative
:-----------------------------------|----------:|--------:
react/no-did-mount-set-state        |   181.362 |     6.2%
react/destructuring-assignment      |   162.586 |     5.5%
react/no-deprecated                 |   134.652 |     4.6%
react/no-direct-mutation-state      |   124.110 |     4.2%
react/boolean-prop-naming           |   116.978 |     4.0%
react/no-did-update-set-state       |   113.393 |     3.9%
react/no-unknown-property           |    92.802 |     3.2%
react/hook-use-state                |    91.712 |     3.1%
react/no-unstable-nested-components |    90.477 |     3.1%
react/no-access-state-in-setstate   |    87.266 |     3.0%

[codecov-commenter]

Codecov Report

Merging #3265 (365849c) into master (c297a96) will decrease coverage by 0.00%.
The diff coverage is 98.00%.

@@            Coverage Diff             @@
##           master    #3265      +/-   ##
==========================================
- Coverage   97.69%   97.69%   -0.01%     
==========================================
  Files         121      121              
  Lines        8586     8583       -3     
  Branches     3118     3120       +2     
==========================================
- Hits         8388     8385       -3     
  Misses        198      198              

Impacted Files	Coverage Δ
lib/util/jsx.js	95.55% <96.96%> (+1.43%)	⬆️
lib/util/ast.js	95.50% <100.00%> (-0.95%)	⬇️

---

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 c297a96...365849c. Read the comment docs.

[golopot]

Avoids unnecessary traversals by skipping all node types but the handled cases.

[golopot]

Same as above and simplify by not using astUtil.traverse.

[ljharb]

I can't pretend I understand this change, but if all the tests pass, then i guess yay?

[golopot]

The gist is that isReturningJSX tries to find return statements that contains JSX. For that a tree traversing is performed. Previously the tree traversing is inefficient because it make unnecessary traversals to nodes that cannot contain a return statement. This PR makes the tree traversal more efficient by only traverse what is needed.