fix: guard parse_multi_exp against missing nodes

[aditya20-b]

Summary

parse_multi_exp assumed every nested key was present. When LinkedIn omits pagedListComponent or a per-role caption / titleV2 (which happens on plenty of multi-role / promoted profiles), it raised TypeError — and the caller's broad try/except swallowed the error and dropped the entire experience entry. The practical effect: members who'd been promoted internally at a company silently lost their tenure in the scraped output, so the people with the longest histories were exactly the ones whose data went missing.

This PR:

• Adds a small _nested() helper that walks a chain of keys and returns None on any miss, instead of throwing.
• Guards every access in the multi-role path so a single malformed role gets skipped rather than torpedoing the whole company entry.
• Falls back to the company-level caption span (the overall tenure) when per-role data is unavailable, so the experience entry is still produced.

No behavioral change for well-formed payloads — only the previously-dropped ones now come through.

Test plan

• Reproduced on a profile that previously returned no experience for a promoted member; now returns the full company tenure.
• Verified single-role profiles still parse identically.
• Verified that a malformed individual role is skipped without affecting the rest of the company's roles.