fix(weekly-digest): truncate activity payload to stay within gpt-4.1-mini's 8000-token limit

[Copilot]

The weekly_digest job was failing with HTTP 413 because gh search returns full nested objects for repository and author, and collects unbounded results — pushing the JSON payload over gpt-4.1-mini's 8000-token request body limit.

Changes

• Flatten nested objects: extract .repository.nameWithOwner and .author.login as plain strings instead of embedding the full gh-search objects (which include id, isPrivate, isFork, etc.)
• Cap result counts: limit issues and PRs to 15 items each, releases to 10
• Drop the events key: was already hardcoded to []; removed entirely

- issues: [.issues[] | {title, number, repository, author, state, url}],
- prs:    [.prs[]    | {title, number, repository, author, state, url}],
- releases: [.releases[] | {name, tag_name, repository, html_url}],
- events: []
+ issues:   [.issues[]   | {title, number, repository: (.repository.nameWithOwner // .repository), author: (.author.login // .author), state, url}] | .[0:15],
+ prs:      [.prs[]      | {title, number, repository: (.repository.nameWithOwner // .repository), author: (.author.login // .author), state, url}] | .[0:15],
+ releases: [.releases[] | {name, tag_name, repository, html_url}] | .[0:10]