Passing skip_rows into scan_csv and read_csv with glob patterns doesn't skip rows before header in first file and doesn't skip rows at all in subsequent files

[qiemem]

Polars version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of Polars.

Issue description

Passing skip_rows into scan_csv/read_csv with glob patterns results in the first line still being used as the header instead of skipping rows the specified number before the header. Also, rows are not skipped at all in any files but the first (whereas the main application of skip_rows is to skip metadata headers that are presumably the same across all files).

For anyone else that runs into this, a workaround is to just do with this manually with the built-in glob library:

pl.concat([pl.read_csv(fn, skip_rows=6) for fn in glob("data_*.csv")])

Reproducible example

Input:

for i in range(3):
    with open(f"test_{i}.csv", "w") as f:
        f.write(
            f"""
metadata goes here
file number {i}
foo,bar,baz
1,2,3
4,5,6
7,8,9
"""
        )

print("With glob")
print(
    pl.read_csv("test_*.csv", skip_rows=2)
    .to_pandas()
    .to_markdown(tablefmt="github", index=False)
)
print("\nSingle file without glob")
print(
    pl.read_csv("test_0.csv", skip_rows=2)
    .to_pandas()
    .to_markdown(tablefmt="github", index=False)
)

Output:

With glob

metadata goes here
foo
1
4
7
metadata goes here
file number 1
foo
1
4
7
metadata goes here
file number 2
foo
1
4
7
metadata goes here
file number 3
foo
1
4
7
metadata goes here
file number 4
foo
1
4
7
metadata goes here
file number 5
foo
1
4
7
metadata goes here
file number 6
foo
1
4
7
metadata goes here
file number 7
foo
1
4
7
metadata goes here
file number 8
foo
1
4
7
metadata goes here
file number 9
foo
1
4
7

Single file without glob

foo	bar	baz
1	2	3
4	5	6
7	8	9

Expected behavior

With glob

foo	bar	baz
1	2	3
4	5	6
7	8	9
1	2	3
4	5	6
7	8	9
1	2	3
4	5	6
7	8	9

Single file without glob

foo	bar	baz
1	2	3
4	5	6
7	8	9

Installed versions

---Version info---
Polars: 0.16.1
Index type: UInt32
Platform: Linux-6.0.16-300.fc37.x86_64-x86_64-with-glibc2.36
Python: 3.9.7 | packaged by conda-forge | (default, Sep 29 2021, 19:20:46) 
[GCC 9.4.0]
---Optional dependencies---
pyarrow: 6.0.1
pandas: 1.4.0
numpy: 1.22.2
fsspec: 2022.01.0
connectorx: 0.2.4
xlsx2csv: <not installed>
deltalake: <not installed>
matplotlib: 3.5.1

[cmdlineluser]

It looks like it's happening here:

https://github.com/pola-rs/polars/blob/master/polars/polars-lazy/src/frame/csv.rs#L305