eds wf job logs: pagination doesn't restore live-tail behavior of removed SSE stream #2

Context

PR #1 replaced the SSE-based eds wf job logs <id> (which streamed lines to
stdout live and stayed open until the job finished — see the removed Long
help text and internal/workflowapi.StreamJobLogs) with a single paginated
POST call (--limit, default 10; --before, default now) against the new
generated client.

Problem

For a running job (the primary "track the rollout" use case documented in
CLAUDE.md), the command now returns at most --limit lines and stops —
there's no loop that pages through to the end of the log, and no live
tailing while the job is still executing. The caller has to notice the
output is incomplete and manually re-invoke with --before to page
further back; there's no way to "wait for more" the way the SSE stream
allowed.

Ask

Either:

• restore a live-tail mode (poll the paginated endpoint on an interval
  until the job reaches a terminal status, printing new lines as they
  arrive), or
• at minimum, auto-paginate to fetch the complete log in one invocation,
  and clearly signal in output when a job is still running and log may
  grow.

Whichever is chosen, update skill/SKILL.md's command table to describe
the actual semantics (see #).

Follow-up from PR #1 review.