Add scheduled good-first-issue candidate digest workflow

[ankitsharma101]

What this does

Adds a scheduled GitHub Action (gfi-candidate-digest.yml) that scans open
issues weekly for heuristic "good first issue" signals and posts a ranked
digest as a new issue for maintainer review. It does not apply the
good first issue label itself — candidates are suggestions only, and a
maintainer decides what to label.

This addresses Section 4.3 of the onboarding/issue-tracking proposal
(follow-up to Discussion #17639 and the June kickoff call), as a concrete,
non-AI, GH Action-based candidate alongside #17901.

How it works

Runs weekly (cron) or on-demand (workflow_dispatch). For each open issue it checks:

• No assignee
• No existing good first issue label
• No triage label (excludes the bot's own prior digests)
• Not opened by a bot account (second safety net against self-reference)
• Updated within the last 6 months (avoids the stale backlog described in Section 2.2)
• ≤3 comments
• Body ≥80 characters

Remaining issues are scored — file/symbol references and repro/expected-behavior
sections in the body nudge score down (better candidate), since issues don't
have a diff to measure scope from directly. The top 10 are posted as a single
digest issue labeled triage.

Known limitation

Issues have no diff, so "small change surface" is approximated by counting
file/symbol references mentioned in the body — this is a weak proxy, not a
real size estimate, and is flagged here intentionally rather than overstated.

Testing

Tested on my fork (ankitsharma101/theia) via manual workflow_dispatch runs:

• Confirmed unassigned, well-described issues get picked up and ranked
• Confirmed assigned issues are excluded
• Confirmed already-labeled good first issue issues are excluded
• Confirmed short-body issues are excluded
• Found and fixed a self-referential bug where the bot's own prior digest
  issues were being treated as candidates in later runs (fixed via bot-author
  and triage-label exclusion)

Still draft pending feedback on thresholds (MAX_AGE_MONTHS, MAX_COMMENTS,
MIN_BODY_LENGTH) tuned against Theia's actual issue volume (~1,440 open),
since I only validated against synthetic test issues on my fork so far.