Finish M2 — subject model + cybermark addresses

state (verified)

The whole mechanism already exists:

• ~self/~now/…/~world lower to axis reads (rs/lower Sym match, hardcoded 2/6/14/…/255)
• #path addresses lower to [17 [path-atom [0 255]]] — scry over ~world
• scry_world walks [[key val] … 0], matches key == path
• hemera hash is real (opcode 15); string_to_atom = hemera::hash(s)[..8] (private in lower)

gaps

1. Subject::minimal() is all Atom(0) → ~self → 0, ~world empty
2. the world-key hash (string_to_atom) is private to the lowerer → nothing else can build matching ~world keys
3. no ~world builder, no populated subject

work

1. rune-subject (+hemera dep):
   • pub fn path_atom(s) -> u64 — the canonical address→world-key hash (hemera)
   • pub fn world(entries: &[(&str, Noun)]) -> Noun — build [[key val] … 0], key = path_atom
   • Subject::standalone() — real-ish: self = neuron atom, here = home, now/world host-filled
   • keep minimal() (zeros) for pure tests
2. rune-lower (+rune-subject dep):
   • string_to_atom → rune_subject::path_atom (one source of truth for the key)
   • hardcoded slot axes → rune_subject::axis::{SELF,NOW,…} (kills magic numbers)
3. cli: a subject() helper — standalone() with ~now = real unix time; used by eval / run / repl instead of minimal()
4. tests:
   • rune-interp integration: build subject with world([("cyber/truth", 42)]), parse+lower+eval #cyber/truth → 42 (scry threading, the completion signal)
   • cli integration: rune eval '~self' → non-zero neuron
   • rune-subject unit: path_atom stable, world layout

completion signal

rune eval '~self' returns a neuron atom; #cyber/truth scry-resolves through a populated ~world. Then update PLAN.md M2 + CHANGELOG.