join_asof breaks with certain parquet files (I think due to memory layout or something?)

[kszlim]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

parquets.zip

Extract these two parquet files:

import polars as pl
l = pl.scan_parquet('left.parquet')
r = pl.scan_parquet('right.parquet')
l.join_asof(r, on='timestamp_monotonic_ns', by='run_id').collect()

Log output

ShapeError: unable to hstack Series of length 409467 and DataFrame of height 448221

Issue description

This throws with:
ShapeError: unable to hstack Series of length 409467 and DataFrame of height 448221

I noticed this is a regression from 0.20.3 -> 0.20.4

Only noticed this after joining some new tables.

Expected behavior

This should work and output something like:

┌────────┬────────────────────────┬────────┐
│ run_id ┆ timestamp_monotonic_ns ┆ status │
│ ---    ┆ ---                    ┆ ---    │
│ i64    ┆ i64                    ┆ bool   │
╞════════╪════════════════════════╪════════╡
│ 1      ┆ 4000000                ┆ null   │
│ 1      ┆ 6000000                ┆ null   │
│ 1      ┆ 8000000                ┆ null   │
│ 1      ┆ 10000000               ┆ null   │
│ …      ┆ …                      ┆ …      │
│ 1      ┆ 896438000000           ┆ false  │
│ 1      ┆ 896440000000           ┆ false  │
│ 1      ┆ 896442000000           ┆ false  │
│ 1      ┆ 896444000000           ┆ false  │
└────────┴────────────────────────┴────────┘

Installed versions

--------Version info---------
Polars:               0.20.31
Index type:           UInt32
Platform:             macOS-14.3.1-arm64-arm-64bit
Python:               3.9.6 (default, Feb  3 2024, 15:58:27)
[Clang 15.0.0 (clang-1500.3.9.4)]

----Optional dependencies----
adbc_driver_manager:  0.5.1
cloudpickle:          <not installed>
connectorx:           0.3.1
deltalake:            0.10.0
fastexcel:            <not installed>
fsspec:               2023.6.0
gevent:               <not installed>
hvplot:               0.9.2
matplotlib:           3.7.2
nest_asyncio:         1.5.6
numpy:                1.26.4
openpyxl:             <not installed>
pandas:               2.2.1
pyarrow:              15.0.0
pydantic:             2.0.2
pyiceberg:            <not installed>
pyxlsb:               <not installed>
sqlalchemy:           2.0.18
torch:                <not installed>
xlsx2csv:             0.8.1
xlsxwriter:           3.1.2```

</details>

[kszlim]

Thanks to @cmdlineluser there's a workaround if within pl.scan_parquet rechunk is set to True.

[kszlim]

Awesome, thanks!