nx_enginelab_store.nx

nx_enginelab_store.nx -- EL1: PERSIST A CAPTURE AND REOPEN IT, for the engine instrument. WHY THIS IS A THIRD MODULE AND NOT PART OF THE SPINE. Same principle that moved golden-image out: nx_enginelab_lib is PURE BUFFER OPS so a wasm world can adopt it, and file I/O is exactly the thing that stops being portable. A consumer that only records frames and zones should not acquire an openat/write/read dependency to do it. So the spine stays pure, and persistence lives here. WHAT EL1 IS FOR. Until now a capture existed only inside the process that took it, which means it could not be attached to a bug report, diffed across machines, or replayed. It also means EL3 (inspect real resource contents per draw) is impossible, because inspecting contents needs a capture that OUTLIVES the process. This is that file. THE LOAD-BEARING REQUIREMENT IS THE REFUSAL, NOT THE ROUND-TRIP. Anyone can write bytes and read them back. The done-rule for this rung is that a TRUNCATED FILE IS REFUSED RATHER THAN PARTIALLY PARSED, because a half-read capture parses into a SHORTER capture that looks completely healthy -- fewer frames, fewer draws, every analyzer happily computing a confident wrong answer over it. That is the silent class this whole board exists to catch, and a persistence layer is where it is easiest to introduce. Hence: a declared byte length, a checksum, and a NAMED reason per failure. AND THE COVERAGE COUNTERS TRAVEL WITH THE DATA. The store header carries drop and error counts, and they are serialised WITH the records. A capture that dropped events under load must not reload claiming complete coverage -- if it did, persistence would launder a partial measurement into a clean-looking one, which is worse than not persisting at all. license_tier: ORIGINAL No hardware writes (Rule 26).