Per-project staleness check in process_projects_hierarchy doesn't gate on requested outputs

From https://github.com/daaain/claude-code-log/pull/272#discussion_r3567034406

Root cause

In claude_code_log/converter.py, the staleness check that decides whether a project needs (re)conversion doesn't distinguish which artifacts were actually requested (write_combined, generate_individual_sessions, use_cache). Today (in _plan_project, and previously in the inline loop it was extracted from):

if write_combined:
    needs_work = (
        bool(modified_files)
        or bool(stale_sessions)
        or combined_stale
        or not output_path.exists()
    )
else:
    needs_work = bool(modified_files) or bool(stale_sessions)

Affected scenarios

1. use_cache=False + write_combined=False (individual-session-only, no cache): cache_manager is None, so modified_files and stale_sessions are always [] (the ternaries fall through to the else branch). needs_work ends up False unconditionally, so no per-session files are ever generated, yet the index still links to session-{id}... paths that were never written — producing 404 links.
2. Combined-only runs (generate_individual_sessions=False, write_combined=True): stale_sessions is still computed and can force needs_work=True (or keep it True) even though individual files aren't being produced and their staleness is irrelevant to what's requested — causing unnecessary reconversion every run.
3. Individual-only runs (generate_individual_sessions=True, write_combined=False): symmetric case — combined_stale/output_path.exists() are correctly skipped, but this branch was only reachable via write_combined, not generate_individual_sessions, so there's no explicit signal tying the check to what was requested; it currently happens to work only because write_combined=False is the only lever, conflating "don't check combined staleness" with "combined wasn't requested."

Proposed fix outline

• Thread generate_individual_sessions into _plan_project (and its caller in process_projects_hierarchy).
• Gate stale_sessions/modified_files evaluation by generate_individual_sessions, and combined_stale/output_path.exists() by write_combined.
• When cache_manager is None (i.e., use_cache=False), don't rely on empty modified_files/stale_sessions lists as "nothing to do" — treat any requested output as needing work whenever there's no cache to have already produced it.
• Add regression tests for: (a) use_cache=False, write_combined=False, generate_individual_sessions=True, (b) write_combined=True, generate_individual_sessions=False steady-state (no unnecessary reconversion), (c) write_combined=False, generate_individual_sessions=True steady-state.

Files/areas

• claude_code_log/converter.py: _plan_project, its call site in process_projects_hierarchy (Phase 1 planning loop).
• test/test_parallel_processing.py or a new staleness-focused test module for the new coverage.

Read claude_code_log/converter.py at _plan_project and its Phase 1 call site in process_projects_hierarchy. Then run the relevant tests in test/test_parallel_processing.py, adding coverage for the three listed cache and output combinations. Done means requested individual outputs are generated without a cache, and steady-state runs do not reconvert outputs that were not requested.