Update all fmt.Errorf occurrences with errors.New using either fmt.Sprintf or strings.Join

[jarri-abidi]

fixes issue #426

[codecov]

Codecov Report

Merging #454 into master will decrease coverage by 0.02%.
The diff coverage is 74.35%.

@@            Coverage Diff             @@
##           master     #454      +/-   ##
==========================================
- Coverage   86.67%   86.64%   -0.03%     
==========================================
  Files          41       41              
  Lines        5104     5108       +4     
==========================================
+ Hits         4424     4426       +2     
- Misses        540      542       +2     
  Partials      140      140              

Impacted Files	Coverage Δ
any_str.go	96.59% <0.00%> (ø)
reflect_map.go	87.86% <0.00%> (ø)
any.go	80.20% <50.00%> (ø)
reflect_array.go	77.58% <50.00%> (ø)
reflect_slice.go	96.61% <50.00%> (ø)
stream_float.go	91.17% <50.00%> (ø)
any_invalid.go	81.08% <66.66%> (ø)
reflect_struct_encoder.go	84.03% <66.66%> (ø)
iter.go	90.09% <100.00%> (ø)
reflect.go	93.00% <100.00%> (+0.14%)	⬆️
... and 2 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 53b9d06...d4b7061. Read the comment docs.

[jarri-abidi]

Could anyone review this?
I'm not sure if tests for these changes would be overkill.
Nevertheless I did add some benchmarks as justification for the changes.

[AllenX2018]

Thanks for the patch. I'm not familiar with fmt package and could you pls explain why using errors.New(fmt.Sprintf()) would decrease the heap usage other than using fmt.Errorf()? Besides, I understand that you add that benchmark test as justification but I don't think its a good idea to put it into json-iterator.It's good enough to just put it in the pr's description.

[jarri-abidi]

So if we take a look at the benchmarks below taken from the added tests, we can see that fmt.Errorf() on its own uses 40 B/op and 3 allocs/op.
errors.New(fmt.Sprintf()) on the other hand uses 24 B/op and 2 allocs/op.
It's also faster in terms of execution time.

152 ns/op 40 B/op 3 allocs/op
119 ns/op 24 B/op 2 allocs/op

In addition to that, where possible, I have replaced fmt.Errorf() with a strings.join() approach that is much faster than either of the above mentioned approaches.