Expose optional GTF parsing progress callbacks for downstream progress bars #80

Reading a large GTF is a long step in pyensembl's genome setup, but gtfparse 2.8.1 exposes no progress callback in read_gtf, parse_gtf, or parse_gtf_and_expand_attributes. A caller cannot show useful parsing progress through its own progress UI.

This is separate from #77: the deprecated Polars string-cache warning was fixed in #78 and published in 2.8.1. Progress reporting remains unimplemented.

Proposed scope:

• Add an optional callback for progress through loading, attribute expansion, and final conversion, with documented units and completion semantics.
• Report measurable work where available (for example, rows processed during attribute expansion); describe opaque operations as stages rather than inventing percentages.
• Keep the default silent and avoid a required progress-bar dependency or global logging configuration.
• Preserve existing output, row order, dtypes, compressed-file support, file-like input support, and filtering behavior.
• Document how callers such as pyensembl can connect the callback to their own progress bars. Implementation in this repository should expose the hook; downstream UI adoption can happen separately.

Acceptance checks:

• Regression tests verify callback ordering, completion, empty/filtered input behavior, and unchanged results when callbacks are enabled or omitted.
• Define and test how exceptions raised by the callback are handled.
• Measure overhead on a representative large GTF, including the default path.
• Run ./lint.sh and ./test.sh, then publish a versioned release.