feat(notebooklm): add core workflows and downloads

[muqiao215]

This PR brings NotebookLM core workflows to opencli and turns the adapter into a usable end-to-end CLI flow instead of a collection of isolated read commands.

Included in this PR

1. Notebook capabilities

• list, current, get, metadata, summary, describe
• create, rename, delete, remove-from-recent
• stable bind-current / use behavior for the active notebook tab

2. Source capabilities

• source list / get / fulltext / guide
• source add-text / add-url / add-file / add-drive
• source rename / delete / refresh / check-freshness
• source wait / wait-for-sources

3. Notes capabilities

• notes list / get / save / create / rename / delete
• explicit --note-id support for unambiguous note operations

4. Notebook interaction

• minimal ask command for the current notebook

5. Generate capabilities

• generate report
• generate audio
• generate slide-deck
• generate infographic
• minimal --wait support on generate commands
• improved generation diagnostics with explicit error_type / message fields when the server rejects a request

6. Download capabilities

• download list
• download report
• download audio
• download video
• download slide-deck
• download infographic
• --artifact-id shell completion for supported download commands

7. CLI compatibility and UX

• compatibility aliases / wrappers aligned with the existing notebooklm-cdp-cli command style
• partial nested command remounts while preserving flat command compatibility

End-to-end workflow now supported

create notebook -> add text/url/file -> wait for sources -> inspect summary/guide/fulltext -> ask -> generate -> download

Known boundaries

• notes-save still requires a visible note editor in the current page. This is a documented design boundary.
• Some download and upload flows depend on temporary signed URLs / resumable upload URLs with TTL. These are runtime constraints, not adapter logic regressions.
• generate report is fully validated end-to-end.
• generate audio, generate slide-deck, and generate infographic are implemented and now return structured diagnostics, but live validation still depends on server-side eligibility / quota / opaque rejection behavior. In current live testing they may return generation_failed_unknown.
• source add-drive is implemented with a narrow input contract (file_id + explicit title + optional mime type) rather than broad Drive URL parsing.
• The adapter relies on real NotebookLM browser state and non-public web RPCs. This PR is intended as a practical automation layer for real user workflows, not a stable official API surface.