Implement project discovery and multi-document lifecycle

[zoeyrose]

Outcome

Open authored content projects safely and manage multiple linked documents without conflating source, generated output, runtime state, or recovery data.

Scope and invariants

• Support explicit open/recent projects, validated roots/manifests, project identity, resource/content version compatibility, and clear diagnostics.
• Manage tabs/documents, active/selected state, dirty markers, linked-map navigation, external changes, close/quit confirmation, reopen, and session restore.
• Keep autosave/recovery outside authored sources with project/file revisions and user-controlled restore/discard.
• Watch source/resource changes with debounce/generation semantics and avoid feedback loops from editor writes.
• Never write generated/runtime/server state paths as authored source.

Acceptance criteria

• Tests cover multiple projects/documents, duplicate opens, rename/delete/external edit, invalid/symlinked roots, crashes, corrupt recovery, reopen, and version mismatch.
• Dirty state is document- and transaction-accurate; closing one document cannot lose another's changes.
• Recovery offers semantic/text diff context before applying and cannot overwrite newer source silently.
• Linked navigation preserves each document's viewport/tool/selection state.
• Large projects meet explicit discovery/watch memory/latency bounds.

Dependencies and parallelization

Depends on #2 and toolkit project/document APIs. UI chrome and toolkit indexing can proceed in parallel.

Licensing

Editor code/tests are MIT. Verified original past work by an approved MIT provenance grantor may be copied, migrated, or translated under the audited file/component grant; other legacy implementation remains behavior/specification input only. Authored content/assets retain exact licenses.