fix(deployless)!: respond in the shape the paged abi declares

[haydenshively]

Follow-up to #49, which shipped paged lenses but had the transport drop half the
lens's declared return value.

The problem

A paged lens declares (U[] results, uint256[] skipped). The transport reassembled
the chunks and handed back a bare U[], discarding skipped. That made paged lenses
unreadable through readContract, decodeFunctionResult, and contract instances —
the entire viem contract layer — because the response no longer matched the fragment:

readContract(client, { abi: [pageAbi], ... })
// Position `129` is out of bounds (`0 < position < 128`)

call2 existed to work around this, handing back { data, missing } from a thrown
error because the real shape had nowhere to go.

The fix

Return the tuple the abi declares. The honest aggregate of N chunk-pages is a page —
results is everything served, skipped is everything that wasn't, rebased from
chunk-local to caller-global indices (and expanded across deduplicated inputs). Nothing
is lost, and readContract works:

const [results, skipped] = await readContract(client, {
  abi: [pageAbi], functionName: 'page', args: [inputs],
  factory, factoryData, address: to,
  stateOverride: [policy({ abi: pageAbi, paged: true })],
})

Unpaged lenses are untouched — no skipped array to report into, so an unservable
element still throws exactly as before.

What this deletes

Everything built to route the dropped array around viem:

Removed	Why it existed
call2	catch the error carrying the dropped array
DeploylessPartialResultError	be that error
TERMINAL_ERROR / isTerminalError	stop failover re-running on it
failover terminal check + telemetry	same
code = -32099	stop viem retrying it

A partial result is now a successful response, so failover never sees it and viem
cannot retry it — both resolved structurally instead of by classification. Net
−368 lines.

Trade-offs, accepted deliberately

A caller who ignores skipped silently gets fewer elements where they'd previously
have caught an exception. The abi forces the array to exist, and re-throwing is one
if at the call site.

skipped merges elements the lens declined with elements that exhausted the frame
even alone. The second depends on the node's eth_call gas cap, so another provider
might serve them, whereas a decline is a property of the element. Called out in the
README and the policy TSDoc.

Verification

492 tests pass, typecheck / build / biome clean. New coverage: readContract against
a paged lens through a real client, the aggregated tuple shape, and the empty-input
page. The two observability tests from #45 that asserted status: "error" on a partial
result now assert "ok" — elements_missing still stamps, and still matches the
skipped array the caller receives.

[haydenshively]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

Reviewed commit: 1804d799e8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[linear-code]

APPS-1267