docs: correct Ethereum network options for roller, claz, and azimuth

[sigilante]

Ninth PR from the audit against urbit/urbit@08026c84b2. Urbit ID network options.

Companion PRs: #266, #267, #268, #269, #270, #271, #272, #273.

The headline for reviewers: these three pages each name a different network type, and they are not the same set. Checking one and generalising gives the wrong answer for the other two — which is exactly the mistake this PR avoids.

Where	Type	Contains %ropsten?
gen/roller/endpoint.hoon	?(%mainnet %goerli %local)	no
sur/claz.hoon +network	?(%mainnet %goerli %fakenet [%other id=@])	no
sur/dice.hoon +$ net	?(%mainnet %ropsten %goerli %local %default)	yes

roller-tutorial.md

:roller|ropsten does not exist — pkg/arvo/gen/roller/ contains goerli.hoon, not ropsten.hoon, and endpoint.hoon declares its own inline type that doesn't accept %ropsten at all. Corrected the command name, the documented argument type, and the intro prose which offered Ropsten as a supported alternative.

Also added two roller generators missing from the command table entirely: |assign [ship (unit @ud)] (per-ship quota; omit the quota for unlimited) and |refuel [@ @ud (unit @ux)] (bump gas price for a sending transaction).

Added a note that the testnet path is unmaintained upstream, not merely in the docs: gen/roller/goerli.hoon hardcodes a Goerli Infura URL, Goerli has been deprecated by the Ethereum Foundation, and there is no Sepolia or Holešky generator in the directory. Readers are pointed at |local plus an explicit |endpoint instead.

advanced-azimuth-tools.md

Said other claz networks are "possible including the test-net %ropsten". +network in /sur/claz.hoon has no %ropsten, and the dispatch in app/claz.hoon:187-192 agrees. Replaced with the real set, noting [%other id] takes a chain ID directly and is the practical testnet choice now.

dojo-tools.md

:azimuth|watch documented its network argument as ?(%mainnet %ropsten %local %default). Here %ropsten is not stale — +$ net in /sur/dice.hoon still includes it. The doc was merely incomplete, missing %goerli, so that is what was added. Removing %ropsten — the obvious move after the other two files — would have been wrong.

flow.md

Used Ropsten as the illustrative example for replay protection. Changed to "a test network" so the example doesn't go stale again; the point is about chain IDs, not any particular chain.

Not touched

Two broken anchors in dojo-tools.md (#vat, #claycancelautocommit ×2) are deliberately left alone — already fixed in #266 and #273 respectively. Verified in a combined test-merge of all nine branches that the merged dojo-tools.md has zero broken anchors and the correct $net type, with each fix applied exactly once.

🤖 Generated with Claude Code