[Feature]: let graphify query start from caller-chosen nodes (--seed) #4293

Problem or use case

Keyword seed selection keeps coming up: #445 (image and doc captions win the seeds), #2381 (test and doc rationale
nodes take over the seeds), #2384 (local variables win the seeds), #3810 (single words from the question cross
domains). Each of those asks for smarter automatic seeding.

Often, though, the caller already knows where to start: an agent that has the file open, a user who knows the symbol,
or an external retriever (embeddings, hybrid search, a learned ranker). Today there is no way to tell query "start
here". You have to word the question so the keyword scorer happens to land there.

Proposed solution

graphify query "<question>" --seed X [--seed Y ...]

• Each value resolves exactly as affected resolves its seed (affected.resolve_seed): node id, unique label,
  callable name, or a repo-relative or absolute file path anchored to the graph's root. There is one interpretation of
  a node reference.
• It fails closed. Values that do not resolve to exactly one node are listed as skipped. If none resolve, the command
  reports it and does not fall back to keyword seeds.
• The header shows Start (given): [...]. Without --seed, nothing changes.

Implemented in #4289 (62 lines and 7 tests; it passes on Python 3.10 and 3.13).

Alternatives considered

• Better automatic seed scoring (the issues above): worthwhile and complementary. --seed is an escape hatch that
  works whatever the scorer does.
• explain X and path A B: they start from a node, but neither gives a question-scoped traversal with --budget
  and --context.
• Configurable seed exclusion (query seeds can be hijacked by test/doc rationale nodes — make seed-path exclusion configurable with tests/ as the default #2381): it fixes one cause, not the case where the caller already knows the start.

Area

Graph building or querying

Are you willing to submit a PR?

Yes, I can work on this (already open: #4289)