[FEATURE] Add get_saved_jobs tool to fetch authenticated user's saved job postings

[alfonzeta]

Feature Description

Add a new read-only MCP tool get_saved_jobs that navigates to the authenticated user's LinkedIn saved jobs page (/my-items/saved-jobs/) and returns the list of saved job postings as raw text plus extracted job IDs.

The tool should follow the same return format as existing scraping tools:
{url, sections: {saved_jobs: raw_text}, job_ids: [...], references?: {...}}

Returned job_ids should be compatible with the existing get_job_details tool so callers can fetch full descriptions in a second step.

Use Case

I built a CV-to-job-offer matching adapter. Today I have to manually copy job descriptions from LinkedIn for each offer I want to evaluate against a CV.

With a saved-jobs tool, my workflow becomes:

1. Browse LinkedIn and save interesting jobs to Saved (native LinkedIn action).
2. Call get_saved_jobs to list saved postings and collect job IDs automatically.
3. Call get_job_details for each ID to pull full descriptions.
4. Feed those descriptions into the CV matching pipeline without manual copy-paste.

This turns LinkedIn's Saved list into a personal job queue that external tools can consume programmatically.

Suggested Approach

Implement as a new tool in tools/job.py, backed by a LinkedInExtractor method (e.g. get_saved_jobs).

1. Navigate to https://www.linkedin.com/my-items/saved-jobs/ using the existing authenticated browser session.
2. Extract innerText from main (same philosophy as search_jobs — minimize DOM dependence).
3. Reuse _extract_job_ids() to collect numeric IDs from a[href*="/jobs/view/"] links.
4. Support pagination via max_pages (1–10, default 3), reusing the search_jobs pagination pattern (scroll sidebar, stop when no new IDs).
5. Return {url, sections: {saved_jobs: text}, job_ids: [...]} plus optional references.

Constraints per project scraping rules:

• Locale-independent: rely on URL patterns and href extraction, not UI text like "Unsave" or "Saved".
• One navigation per tool call (single section = single page target).
• Verify end-to-end against live LinkedIn with uv run before merging.

Out of scope for v1 (can be follow-up issues):

• Saved vs Archived tab filtering
• Mutations (unsave, apply, change stage)