Improve test coverage

[WestonVoglesonger]

The suite already covers the core surfaces (canonicalize ×3, checkers, create-target, predict, sampling, options), and test-coverage.yaml already runs covr and uploads to Codecov in CI. But some surfaces are thin or untested. This issue tracks closing those gaps as part of the CRAN push (#31).

Gaps to close

• summary.imugap_predict — added with the draws-compatible predict refactor (adjust predict to return a more-draws-compatible object #68 / PR change predict returns to be draws-compatible #70); no dedicated test. Cover the reduction/summary output shape and columns.
• predict.imugap_fit edge cases — beyond the happy path: invalid loc_id / dose / age / cohort rejection, single-row vs. multi-row targets, the different create_target modes feeding predict.
• checkers.R — the checked_* validators throw on bad input; assert the error paths (each stopper branch), not just the pass-through.
• options.R (stan_options, imugap_options) — default construction + override + invalid-arg handling.
• canonicalize_* — confirm the failure/validation branches are exercised, not just valid inputs.

Acceptance

• summary.imugap_predict has tests
• checker/validator error branches covered
• options + canonicalize failure paths covered
• predict edge cases covered

Refs #31