[REVIEW] Explicit dictionary constructor

[bdice]

This PR makes the dictionary32 constructor from an int32 explicit. An implicit constructor here defeats the purpose of being a strongly typed wrapper.

This is a dependency of #5494.

The underlying issue this solves is similar to a change made in #5735, where timestamps could be constructed from a duration type's representation. Quoting from discussion with @jrhemstad:

The exact mechanism that caused this operator* problem is this:
Boiling away the cruft, the operator* definition looks like this:

template <class T, class Period, class Rep>
enable_if_t
<
    is_convertible<T, common_type_t<T, Rep>::type>::value,
    duration<common_type_t<T, Rep>::type, Period>
>
operator*(const T& s, const duration<Rep, Period>& d);

So this allows multiplication between a duration type and any type T so long as T is convertible to the common type between T and the duration's representation Rep. "common_type" is defined as "For a list of types, what type are they all implicitly convertible to?" e.g., common_type_t<int, short, char> == int
So for example, std::chrono::days::Rep == int32_t.
If T == timestamp_D, then it first checks "What is the common type between timestamp_D and int32_t?". Well, because int32_t is implicitly convertible to timestamp_D (thanks to that constructor) then common_type_t<timestamp_D, int32_t> == timestamp_D. So then we end up with is_convertible<timestamp_D, timestamp_D>, and that is true because a type is always convertible to itself.
Therefore, the type substitution for operator* passes, even though the actual instantiation of operator*(timestamp_D, duration<...>) is ill-formed.

In this case, the implicit constructor of a dictionary32 from an int32 meant that it met the criteria for the overload of operator* with a duration, which isn't desired.

Aside from the change to the constructor, this only affected one place that relied on the implicit constructor.

[GPUtester]

Please update the changelog in order to start CI tests.

View the gpuCI docs here.

[codecov]

Codecov Report

Merging #5832 into branch-0.15 will increase coverage by 0.32%.
The diff coverage is n/a.

@@               Coverage Diff               @@
##           branch-0.15    #5832      +/-   ##
===============================================
+ Coverage        84.08%   84.40%   +0.32%     
===============================================
  Files               80       80              
  Lines            13061    13381     +320     
===============================================
+ Hits             10982    11294     +312     
- Misses            2079     2087       +8     

Impacted Files	Coverage Δ
python/cudf/cudf/io/feather.py	100.00% <0.00%> (ø)
python/cudf/cudf/comm/serialize.py	0.00% <0.00%> (ø)
python/custreamz/custreamz/_version.py	0.00% <0.00%> (ø)
python/dask_cudf/dask_cudf/_version.py	0.00% <0.00%> (ø)
python/dask_cudf/dask_cudf/io/tests/test_orc.py	100.00% <0.00%> (ø)
python/dask_cudf/dask_cudf/io/tests/test_json.py	100.00% <0.00%> (ø)
...ython/dask_cudf/dask_cudf/io/tests/test_parquet.py	100.00% <0.00%> (ø)
python/cudf/cudf/utils/applyutils.py	98.75% <0.00%> (+0.02%)	⬆️
python/cudf/cudf/core/join/join.py	92.41% <0.00%> (+0.03%)	⬆️
...ython/custreamz/custreamz/tests/test_dataframes.py	98.80% <0.00%> (+0.03%)	⬆️
... and 29 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 f252a93...9344cc4. Read the comment docs.