Enable bias by default for linear models

[Same0620]

Summary

• Use 1.0 as the default bias when -B is omitted.
• Add the bias feature column after resolving the effective bias value.
• Preserve -B -1 as an explicit way to disable the bias term.

Motivation

The previous default of -1.0 disabled the bias term when users did not provide -B. Simply changing the default value was insufficient because the bias feature column was only added inside the explicit -B branch. Moving that check after option parsing makes the default behavior equivalent to passing -B 1 while preserving existing explicit settings.

Validation

• python -m black --check libmultilabel/linear/linear.py
• Ran a local smoke test covering default, explicit, and disabled bias settings, including one-vs-rest training and prediction.

[Same0620]

Superseded by #31, which uses a fork-based branch.