Quick fix for #14815

this was bothering me on startup :)

See `#14137`, I think there are invisible completions and when the
current line is a prefix of an existing history entry, we happend to
have autocompletions but they are not shown.

I need to check how this affects the multiline completions though, and
we might need a more general solution.

This otherwise break pycharm

See #14810

This should fix issue #14825

This should fix issue #14825

fixes: #14818

The `leading_indent` function in IPython/core/inputtransformer2.py had
an issue in that it checked to see if the first line had a comment
anywhere in it, and if it did, then it just would ignore checking if the
rest of the input could possibly need to have indentation removed.

The main issue is that it looks like the function assumes you can just
check the first line, but that is not the case. You need to check the
indentation level of every line in the input and remove the minimum (in
terms of length) indentation common to all lines.

I'm not sure why this function isn't just using `textwrap.dedent`, which
seems like it does exactly what we want to be doing. I just switched the
function to use this.

For the tests, it turns out that there was a test cases that wasn't even
being checked. I fixed that as well and added the sample I noted in
#14818 as a 3rd test case.

While working on this, I wrote a more involved test, which didn't really
fit with the style of the other tests, but I'll put it here in case you
want to include it.

<details>

```python
def test_regression_14818():
    from IPython.core import inputtransformer2 as ipt2
    from textwrap import dedent

    def indent(text, prefix='    '):
        """Indents a block of text"""
        return prefix + text.replace('\n', '\n' + prefix)

    targets = {}
    targets['two_statements'] = dedent(
        """
        x = 1
        print(x)
        """
    ).strip()
    targets['aligned_comment_and_two_statements'] = dedent(
        """
        # This is a comment
        x = 1
        print(x)
        """
    ).strip()
    targets['misaligned_comment_and_two_statements'] = dedent(
        """
            # This is a misaligned comment
        x = 1
        print(x)
        """
    ).strip()

    # Check that all of the targets are valid python
    for target_name, target_code in targets.items():
        exec(target_code, {}, {})

    # Modify the target text to test that ipt2.leading_indent can undo the
    # modifications.
    cases = []
    for target_name, target_code in targets.items():
        cases.append({
            'name': target_name + '_space',
            'sample': indent(target_code, prefix='    '),
            'target': target_code,
        })
        cases.append({
            'name': target_name + '_tab',
            'sample': indent(target_code, prefix='\t'),
            'target': target_code,
        })

    for case in cases:
        case_name = case['name']
        lines = case['sample'].splitlines(keepends=True)
        recon_lines = ipt2.leading_indent(lines)
        reconstructed = ''.join(recon_lines)
        assert reconstructed == case['target'], f'failed case {case_name}'
```

</details>

This otherwise break pycharm

Plus a few other fixes.

See #14810

See `#14137`, I think there are invisible completions and when the
current line is a prefix of an existing history entry, we happend to
have autocompletions but they are not shown.

I need to check how this affects the multiline completions though, and
we might need a more general solution.

Quick fix for #14815