feat(sync): GBRAIN_NO_PRUNE_DIRS — un-prune a content dir (e.g. an ops/ KB)

[donogeme]

What

Adds GBRAIN_NO_PRUNE_DIRS (comma-separated env var) that removes directory names from PRUNE_DIR_NAMES at walk time, so a brain can sync a tree whose top-level name happens to collide with a pruned name.

Why

PRUNE_DIR_NAMES = {node_modules, .raw, ops} is hardcoded. ops in particular is a semantic prune ("operations, not knowledge") — great for repos that use ops/ for throwaway operational files. But some brains use ops/ as a genuine knowledge base (e.g. ops/_kb/{compliance,finance,partners,...}). For them, the entire tree is silently invisible to search with no opt-out — a single hardcoded name decides it.

This gives an escape hatch without changing any default: set GBRAIN_NO_PRUNE_DIRS=ops and that tree syncs.

Behavior

• Comma-separated; trims whitespace, drops empties/dupes.
• Only un-prunes names in PRUNE_DIR_NAMES (and the *.raw suffix rule).
• Dot-prefixed dirs (.git, .obsidian, …) and git submodules stay pruned regardless — those are never content.
• Unset/empty = exactly today's behavior. Purely additive, no default change.
• Routes through pruneDir, so every walker (sync, extract, transcript-discovery) honors it consistently.

Example

GBRAIN_NO_PRUNE_DIRS=ops gbrain sync --all

Tests

test/sync-no-prune-dirs.test.ts — parser contract + prune/un-prune matrix (ops un-pruned, node_modules still pruned, dot-dirs always pruned). bun run typecheck clean.

Context

Env-var form mirrors the existing GBRAIN_SOURCE/GBRAIN_SOURCES convention. Happy to reshape as a sync.* config key instead if you'd prefer config-driven over env.