fix: unreachable chromid warning, shared mutable defaults, unclosed bed file handles

[sanjaynagi-eit]

Three unrelated correctness issues found while profiling. None are performance changes; opening them separately so the perf PRs stay reviewable.

1. The multiple-chromosome warning could never fire

c = 1
...
if c == 1:
    dna_header = "chromosome"
else:
    if c == "2":   # <- c is an int
        logger.info("Multiple contigs above the specified chromosome length -c have been detected. ...")

c is an integer, so c == "2" is always False. A user assembling something with two chromosome-sized contigs — chromids, or simply -c set too low — silently got no hint that this had happened, which is exactly the case the message exists for. Three copies, in the raven, flye and flye_long variants.

2. Mutable and import-time-evaluated default arguments

depth_df: pd.DataFrame() = pd.DataFrame({"col1": [1, 2, 3], "col2": [4, 5, 6]}),
filtered_out_contig_ids: list = [],
plasmid_names: list = ["1"],

• The list defaults are created once at import and shared by every instance. Currently benign because both are always reassigned before use, but it is a trap waiting for the first .append().
• pd.DataFrame() in annotation position is not a type — it constructs and discards a DataFrame at import time, six times across the two classes.
• The default value was a dummy {"col1": [1,2,3], "col2": [4,5,6]} frame rather than an empty one, so an un-populated Plass carried fake data.

All now default to None and are materialised per instance.

Separately, Assembly.__init__ accepted chromosome_name and plasmid_names and then never stored them. It does now.

3. BED file handles were never closed

bed_file = open(os.path.join(outdir, "non_chromosome.bed"), "w")
bed_chrom_file = open(os.path.join(outdir, "chromosome.bed"), "w")

Bare open(), no with, no .close(). samtools view -L reads both files later in the run, which works today only because CPython's refcounting happens to flush and close them when the method returns — it would break under a non-refcounting implementation, and it makes the ordering guarantee invisible. Closed explicitly.

Tests

New tests/test_plass_class_defaults.py: two instances no longer share default containers, explicitly supplied values still win, the DataFrame defaults are empty rather than dummy data, and the multiple-chromosome warning actually fires for an assembly with two chromosome-sized contigs.

All existing tests pass (74 non-slow).