Correct release qualification boundaries

[raghubetina]

No description provided.

[raghubetina]

Post-merge review, hosted CI green; docs-and-tests only. Three precision corrections to release evidence, each
replacing a loose claim with an exactly bounded one: the product-journey smoke now names the historical CLI
revision and version it actually drove (f55edffc, 0.1.0-alpha.2) instead of implying the current successor;
the catalog revision is labeled as a catalog-promotion revision rather than "this exact main," which drifts
the moment main moves; and the Compilation/Publication account now distinguishes the separately billed Render
One-Off that ran the Compile from the persistent Solid Queue worker that ran Publication coordination, which
are different execution paths with different failure modes. The 0.1.0 teaching-surface limitation is stated as
an explicitly accepted documentation limitation rather than silently patched. All four changed files agree.
This is the unglamorous work that makes the later promotion records (#31 through #33) trustworthy.

[raghubetina]

Every claim in this diff got smaller, and that is the point. Three edits show what evidence hygiene looks like
in practice.

"The smoke drives the pinned successor CLI" became "the smoke drove exact revision f55edffc, version
0.1.0-alpha.2." The first phrasing silently upgrades old evidence every time the successor changes; the second
is true forever. Any sentence whose truth depends on when you read it is a bug in a record.

"The public catalog at this exact main revision" became "at catalog-promotion revision." Same hash, better
noun. main is a moving pointer, so describing evidence by its relationship to main builds in decay;
describing it by the event (the promotion) and its hash does not.

The third edit is the most Rails-relevant: the record now says the Compile ran in a separately billed Render
One-Off while Publication coordination ran in the persistent Solid Queue worker. Before, "the service compiled
and published" blurred two processes with different lifecycles, billing, and failure modes. When something
breaks at 2am, "which process was that in" is the first question, and evidence that pre-answers it is worth
the extra clause.

The meta-lesson: after any impressive result, reread your own write-up hunting for claims that are slightly
larger than what you observed. Shrinking them costs a sentence now and buys you a record nobody has to
distrust later.