Possible VCF scanner corruption

[mwiewior]

So far, I find some strange behavior from time to time but I do not know if it is a bug or not. For example, today I had some issue with repeated scans:
The polars_bio VCF scanner has a bug where after ~4 scans of the same LazyFrame, the underlying Rust/DataFusion execution plan's file handle or state becomes corrupted, and subsequent scans return empty results (with a Rust panic).

[mwiewior]

1. Repeated same LazyFrame 10 times with a small collect:

• lf = pb.scan_vcf(path)
• lf.select(["chrom", "start"]).limit(10).collect()
• Result: all 10 iterations returned rows=10, no panic.

2. Repeated full scans of the same LazyFrame 6 times:

• lf = pb.scan_vcf(path)
• lf.select(["chrom", "start"]).collect()
• Result: all 6 iterations returned rows=86821100, ~21s each, no empty result, no Rust panic.

So I could not reproduce issue #313 in this environment on the current code.
If you want, I can try reproducing with a more exact workload pattern (same filters/selections/options you used when it failed).

[mwiewior]

This happens because you called .collect() first (which streams data and deregisters the tables) ???