feat(session): add lifecycle management — storage reclamation, CLI commands, VACUUM support

[guazi04]

Issue for this PR

Partially addresses #16101 (MVP path: steps 1–4)

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Sessions currently have no lifecycle beyond creation. This PR adds the missing pieces from the MVP path outlined in #16101:

Storage reclamation in compaction — prune() in compaction.ts previously marked time.compacted on tool parts but never cleared the heavy output and attachments fields. Now it actually clears them (set output to [reclaimed], attachments to []). A retroactive pass in process() catches parts that were marked compacted before this change but still carry stale content. Gated by a new compaction.reclaim config option (default: true). This implements the conservative reclamation level described in #16101.

CLI session management — Four new subcommands under opencode session:

• archive <id> / unarchive <id> — surfaces the existing setArchived() API that had no CLI exposure
• stats — shows DB/WAL file sizes, session counts (root/child/archived), message/part counts via single-query SQL aggregates
• prune — bulk cleanup with --older-than (default 30 days), --children, --vacuum, --dry-run. Uses direct Drizzle queries with batch pagination to avoid listGlobal() ordering bias. Tracks descendants via Set<string> to prevent double-delete.

VACUUM support — Database.vacuum() does WAL checkpoint (TRUNCATE) + VACUUM; Database.checkpoint() does checkpoint only. Used by session prune --vacuum.

Data integrity fix — remove() previously called children() which filters by project, causing cross-project child sessions to be orphaned on delete. Added childrenAll() that queries by parent_id only, and remove() now uses it for safe cascade deletion.

Not included (future work per #16101):

• Auto-archival for inactive sessions (proposal 2)
• Aggressive reclamation mode (proposal 1, level 2)
• Child session retention period (proposal 3)
• TUI integration for archive/stats (proposal 4, TUI portion)

How did you verify your code works?

• Ran opencode session stats against a real vault with ~200 sessions — counts match manual spot-checks
• Ran opencode session prune --dry-run --older-than 7 to verify candidate selection, then without --dry-run to confirm deletion + vacuum reclaimed ~40 MB
• Tested archive / unarchive round-trip on multiple sessions
• Verified compaction reclamation by inspecting part records before/after compaction — output fields correctly replaced with [reclaimed]
• Confirmed childrenAll() finds children across projects by creating a child session with a different project context

Screenshots / recordings

N/A — CLI-only changes, no UI modifications.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR