fix(mcp): resolve projects by internal name + filter ghost dbs (#704)

[DeusData]

Fixes #704.

Symptom: query_graph / search_graph / get_architecture return "project not found or not indexed" for a project that list_projects shows — inconsistently ("works now, fails 10 min later"). (_config.db being empty is a red herring — it's a KV config store, never a registry.)

Root cause: resolution is registry-less + filename-addressed (list_projects and the error's available_projects come from .db filenames) but a query requires the internal projects.name row to equal the passed name. When a db's filename ≠ its internal name — a copied/renamed db, or a legacy .-vs-- username twin from a past path sanitization — it lists but never resolves. The intermittency was a warm in-process store cache (60 s idle evict) masking a broken cold resolve; a fresh cli call is always cold. Node rows are keyed on the internal name, so opening the file isn't enough — queries must use the internal name too.

Fix — key list + resolve on the INTERNAL name; no file renames, no data mutation:

• list_projects / collect_db_project_names / build_project_json_entry advertise each db's internal projects.name and skip ghost dbs (0-byte / no project row), so listed names are exactly the resolvable ones.
• resolve_store keeps the fast path (open <passed>.db + get_project — the common case). On a miss it runs a bounded cache-dir scan, adopting the db whose sole internal project name equals the passed name (gated on exactly one project row, so it never serves another project's data). No match → not-found (typos stay not-found). The scan runs only on the fallback.

Reproduce-first: tool_resolve_store_by_internal_name_issue704 — isolated CBM_CACHE_DIR with a control db, a drifted db (filename ≠ internal name), and a 0-byte ghost; asserts list advertises the internal name (not the filename), the drifted project resolves via the scan, and the ghost is neither listed nor resolvable. RED (113/1 — list lacked the internal name) → GREEN (114/0). Full suite 5723/0.

Also hardened tool_bad_project_name_no_overflow_issue235: its fixture used 1-byte non-SQLite files that the new ghost filter would skip, making the buffer-overflow guard vacuous — rebuilt with valid dbs carrying long internal names so the guard still fires.

Design note: chose internal-name addressing over a file-rename migration — it makes list/resolve/queries consistent without renaming or mutating any user .db file. The separate latent READWRITE/WAL query-open bug (queries mutate dbs / fail on a read-only FS) is tracked as its own PR.

Part of the CLI-reliability series: #640 → #714, #680 → #716.