CLI silently drops non-PASS records and SVs; --variant chr contigs fail with GRCh38

The CLI silently drops non-PASS records and structural variants, and its own --variant help example fails with --genome GRCh38.

Tested on main at 50eafde (8.0.0), with the #429 import fix applied.

1. Non-PASS records are dropped without notice. load_vcf defaults to only_passing=True, and the CLI has no flag to change that or report it:

   $ varcode-genes --vcf tests/data/mutect-example.vcf --output-csv out.csv
   $ wc -l out.csv
   11            # the VCF has 4889 records: 10 PASS, 4879 REJECT
   

   Expected: an --include-filtered flag, plus a stderr note such as Skipped 4879 records with FILTER != PASS.

2. Structural variants can't be annotated from the CLI. parse_structural_variants isn't exposed. The run exits 0 with an empty table, and the warning suggests a Python argument:

   $ varcode --vcf tests/data/osteosarc_esvee_somatic.vcf --genome GRCh38 --output-csv sv.csv
   .../varcode/vcf.py:454: UserWarning: Skipped 8 symbolic/breakend allele(s) in .../osteosarc_esvee_somatic.vcf (e.g. <DEL>, <CN0>, <INS:ME:ALU>, breakends). Pass parse_structural_variants=True to load_vcf to load these as StructuralVariant objects.
   <EffectCollection with 0 elements>
   $ cat sv.csv
   variant,contig,start,ref,alt,is_snv,is_transversion,is_transition,gene_id,gene_name,transcript_id,transcript_name,effect_type,effect
   

   The same happens for tests/data/spec_examples/vcf43_spec_sv.vcf. From Python, load_vcf(..., parse_structural_variants=True).effects() returns 68 effects for the esvee file (GeneFusion for KIF3C, OTX1, PARM1, SEMA6A and others). Expected: a --structural-variants flag, or SV parsing on by default now that the default annotator routes SVs, and a warning that names the CLI flag.

3. --variant with chr contigs fails for GRCh37/GRCh38 but works for hg19/hg38. The help text's own example is chr1 3848 C G:

   $ varcode --genome GRCh38 --variant chr12 25245350 C T
   ValueError: Invalid contig name 'chr12' for reference 'GRCh38'
   $ varcode --genome hg38 --variant chr12 25245350 C T
   ... KRAS-201 p.G12D
   

   Variant only strips chr when the genome name is a UCSC alias, while load_vcf converts it for any genome. The CLI could pass convert_ucsc_contig_names=True, as the VCF path does. The error also appears only after the sequence dictionaries load, not at argument parsing.

4. One bad record aborts the whole file. A single reference mismatch or unknown contig raises out of effects(). The error text says "Pass raise_on_error=False to .effects()", which a CLI user can't do. Expected: a --skip-errors flag that maps to raise_on_error=False and reports failures as Failure rows.

5. Help text.

   • Typos: "chromsome", "specfied".
   • --vcf, --maf, --variant and --json-variants can be repeated, but the help doesn't say so.
   • --genome lists mm9 as an example, but varcode --genome mm9 --variant 1 3000126 G T fails with Failed to infer genome assembly name for mm9. mm10 and GRCm38 work.
   • --genome can't select a release (#214).
   • --output-csv doesn't say what the columns are.

Start with the CLI entry points and vcf.py's load_vcf and Variant paths, then reproduce the commands using tests/data/mutect-example.vcf, tests/data/osteosarc_esvee_somatic.vcf, and tests/data/spec_examples/vcf43_spec_sv.vcf. Trace how effects() handles filtered records, structural variants, contigs, and errors. Done means the requested flags, diagnostics, contig behavior, and help corrections work without silently losing records.