[v0.5] Generate survival variable pairs inside create_mock_data()

[DougManuel]

Part of the post-v0.4.0 development plan (development/post-v040-development-plan.md, Phase 2, Task 7).

Context

Longest-standing known issue (NEWS, since v0.2.0): survival data must be generated manually with create_wide_survival_data() because it produces paired entry/event columns, while create_mock_data() assembles one column per variable.

Metadata is already plumbed: followup_min, followup_max, event_prop flow through the recodeflow adapter onto spec variables (R/mock_spec_recodeflow.R:367-369); rate/shape likewise (:365-366). The blocker is architectural, not statistical.

Design decisions (small design note, then implement)

1. Pairing convention in metadata — how an event variable names its entry/anchor variable. Plan recommendation: an explicit anchor extension column on the event variable's row.
2. Multi-column generator contract — a spec variable of type survival whose generator returns a 2-column data frame; assembly loop accepts multi-column returns; validate_mock_spec() checks pairing integrity (anchor exists, is a date variable, ordering constraints).
3. Diagnostics attribution — both columns attributed to the one spec entry in mockdata_diagnostics.

Acceptance

• create_mock_data() generates entry/event pairs from metadata alone (no manual create_wide_survival_data() call).
• Event/censoring proportions tested with two-sided bounds (fulfils Add survival censoring and LTFU regression tests #23's direction).
• Known-issue entry removed from NEWS with a migration note.

Relates to #17, #23.