fix(ci): pass the crates.io token to FerrFlow in reusable-ferrflow-release

[BryanFRD]

Closes #355

Two lines. reusable-ferrflow-release.yml exported only the Kellnr credential, so any repo declaring a publisher on the default registry ran cargo publish with no token and failed on "no token found". Every repo releasing through this workflow could publish to Kellnr and nowhere else.

         env:
           CARGO_REGISTRIES_KELLNR_TOKEN: ${{ secrets.CARGO_REGISTRIES_KELLNR_TOKEN }}
+          CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_FERRLABS_REGISTRY_TOKEN }}

Applied to both jobs. publish covers deferred publishers, release covers repos on deferPublish: false that publish inline.

Why the names do not match

CARGO_REGISTRY_TOKEN and CARGO_FERRLABS_REGISTRY_TOKEN both exist as org secrets with visibility=all, and the one that actually publishes to crates.io is the second. FerrLabs/LFSX is the proof: its own publish.yml maps exactly this pair, and lfsx sits on crates.io at 1.17.0. LFSX works today because it calls cargo publish itself instead of going through this workflow.

The left side of the mapping is the env var cargo reads, the right side is where the value lives. Wiring the obvious secrets.CARGO_REGISTRY_TOKEN instead would look correct and publish nothing, which is the trap this removes.

The secret is not declared in the secrets: block, matching the Kellnr line directly above it. Callers use secrets: inherit, which already grants access, and the existing Kellnr publishes prove that path works.

Not verified end to end

The only real proof is a crates.io publish, and that needs a release to fire. FerrLabs/Kit#295 moved four crates onto the default registry, so the next feat: or fix: touching one of them is the first live run. Kellnr publishes are unaffected either way, since that line is untouched.

[ferrfleet]

Change is correct and additive: both jobs run FerrFlow, both can reach cargo publish (inline for deferPublish: false, in the publish job otherwise), so both need the token.

Checked the callers, since the change relies on secrets: inherit reaching an undeclared secret. Five of the six inherit. FerrLabs/FerrVault passes secrets explicitly (FERRLABS_DISPATCH_TOKEN only), so neither cargo token reaches it, but its .ferrflow declares only docker and helm publishers and it is pinned to an older SHA, so nothing breaks there. Worth knowing if a non-inheriting caller ever gains a cargo publisher: the token arrives empty and the failure will look like an auth error, not a wiring error.

One nit inline. Not blocking.

[ferrfleet]

Nit: the name mismatch is explained in the PR description but not in the file. Someone reading just the YAML sees CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_FERRLABS_REGISTRY_TOKEN }}, assumes a typo, "fixes" it to secrets.CARGO_REGISTRY_TOKEN, and publishing silently breaks again. The Kellnr line above carries a comment; this one should too.

Suggested change

	CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_FERRLABS_REGISTRY_TOKEN }}
	# crates.io credential. The names deliberately differ: the org secret
	# that actually publishes to crates.io is CARGO_FERRLABS_REGISTRY_TOKEN,
	# not the identically-named CARGO_REGISTRY_TOKEN. See #355.
	CARGO_REGISTRY_TOKEN: ${{ secrets.CARGO_FERRLABS_REGISTRY_TOKEN }}