Make the POST cognitive drill how-to cards reachable after the first run

[atomantic]

Problem

The POST cognitive drills show a first-run how-to card (DrillTutorialGate in client/src/components/meatspace/post/PostCognitiveDrillRunner.jsx), gated on a per-type flag in localStorage (portos.post.drillTutorialSeen). It is one-shot per drill type, forever — the card itself says "You'll only see this the first time for each drill type."

That is right for the flash-of-stimulus problem it was built for, but it leaves no way back. The n-back card now carries a worked example strip (the letter stream with the one match called out) precisely because the rule is easy to misread — and after the first run that example is unreachable. A user who last did n-back weeks ago gets no refresher, and there is no way to preview a drill type without starting a scored run.

Approach

Add a "How it works" affordance to the drill screen itself, next to the DrillHeader label. Tapping it re-shows CognitiveDrillTutorial for the current type.

Decision on the mechanic (do not re-litigate): do not re-open the tutorial mid-run. The runners are timed and startedAtRef is stamped at mount, so unmounting a live drill would either lose the run or record a bogus totalMs. Instead, show the affordance only in the pre-roll window — n-back's 800ms Get ready… phase and each runner's equivalent — or, simpler and preferred, put the entry point on the launcher (PostSessionLauncher.jsx / PostDrillConfig.jsx) as a per-drill-type "preview how it works" that renders the same CognitiveDrillTutorial with no runner behind it. That keeps the timed path untouched.

getDrillTutorial(drill) is already a pure data function and CognitiveDrillTutorial already renders standalone, so both are reusable as-is; the drill payload needed for config-dependent copy (n-back lag, digit-span direction, reaction-time mode) is available at config time.

Acceptance criteria

• A user can view any cognitive drill type's how-to card, including the n-back worked example, without starting a scored run.
• The first-run gate keeps working unchanged: still one-shot per type, still holds the runner's timers until Start.
• No path re-mounts a live runner, and no drill result records a totalMs that includes time spent reading the tutorial.
• Test coverage for the new entry point alongside the existing "first-run tutorial gate" block in PostCognitiveDrillRunner.test.jsx.

[atomantic]

Implementing the launcher/config entry point (the issue's preferred option), placed on both surfaces named in the Approach:

• POST launcher sidebar — a book icon per enabled cognitive drill row, opening the how-to card against that drill's live config. This is the discoverable one: it sits on the screen you launch a session from.
• Drill Config cognitive cards — a "How it works" link on every card, including drills that are currently switched off, so the acceptance criterion "any cognitive drill type" is actually covered (the launcher only lists enabled drills). It also lets you read the rules before deciding to enable a drill.

Both render the same CognitiveDrillTutorial body through a new CognitiveDrillTutorialPreview modal with no runner behind it, so no startedAtRef is stamped and nothing is timed or scored. The live drill path is untouched.