PipelineParameter parses my timestamp strings

I have the following PipelineParameters
```python
train_start_time = PipelineParameter(
name="train_start_time", default_value="2021-01-01T00:00:00"
)
train_end_time = PipelineParameter(
name="train_end_time", default_value="2021-01-31T00:00:00"
)
```
which are passed to my PythonScriptStep as follows:
```python
pipeline_step = PythonScriptStep(
..
arguments=[
"--train_start_time",
train_start_time,
"--train_end_time",
train_end_time,
],
...
)
```
But when I run this in a compute cluster it turns out that the timestamp strings have been parsed, and is passed on to the script in a different format! In `70_driver_log.txt` I see this:
```
[2021-03-23T14:30:23.703839] After variable expansion, calling script [...] with arguments:['--train_start_time', '01/01/2021 00:00:00 +00:00', '--train_end_time', '01/31/2021 00:00:00 +00:00']
```
Iso8601 is the only way to represent time as strings, any other way is abominations! How do I get my pretty strings?

**Edit for extra clarity**:
The problem is that my input string `2021-01-31T00:00:00` gets transformed to `01/31/2021 00:00:00 +00:00` before beeing passed to my script as arguments.

使用所示的時間戳值重現 PipelineParameter 和 PythonScriptStep 範例，然後檢查 70_driver_log.txt 中變數展開後的引數。比較原始的 ISO8601 字串與轉換後的值，並追蹤該轉換發生的位置。當腳本接收到未變更的原始時間戳字串時，即視為完成。

由索引模型根據 Issue 內容生成。