fix(oneinch): resolve AR v6 trace-decoded token details via calldata fallback

[bh2smith]

Description:

Fix: 1inch AR v6 trace-decoded trades ship with empty-string symbols and NULL amounts (~10%/day)

Problem. dex_aggregator.trades rows for project = '1inch', version = 'AR v6' that come from the trace-decoding path (evt_index = -2) systematically ship with '' (empty-string) token symbols and NULL token amounts: 22,816 rows June 1–12 2026 (8–13% of AR v6 rows per day, 98% on bnb), carrying ~$4.6M/week of measured amount_usd on rows with no token detail. Because the symbols are '' rather than NULL, IS NULL-based quality checks report these rows as 0% incomplete.

• Daily blank rates: https://dune.com/queries/7709025
• Side-by-side repro (spell row blank vs the same fill fully detailed in pool-level data): https://dune.com/queries/7709014 — anchor tx (bnb) 0x463b95a76eb82b6fda8e8a51f96291fef9714c550f9de17f14a37544a89ecfa9

Root cause. The AR executions macro resolves token metadata and amounts from the transfer graph around the router call. When the router is invoked internally (e.g. wallet-proxy → AR v6 0x111111125421ca6dc452d289314280a0f8842a65, selector 0x07ed2379) and/or the fill settles through PancakeSwap-Infinity / v4-style flash-accounting vaults, the transfer-matching filters find no rows, so src/dst amounts and symbols resolve to nothing — even though the swap's token addresses and amounts are present in the decoded call parameters. The trades model then coalesced unresolved symbols to ''.

Fix (two files, no restructuring):

1. oneinch_ar_executions_macro.sql — add calldata-decoded amounts as the final fallback in the existing coalesce chains, and a LEFT JOIN to the tokens lookup on the calldata token address (native sentinels 0x0…0/0xeee…e mapped via blockchain.wrapped_native_token_address / native_token_symbol) so symbol/decimals resolve when transfer-based resolution is empty. Transfer-derived values still take precedence — event-decoded AR v4/v5 output is unchanged.
2. oneinch_ar_trades.sql — stop coalescing unresolved symbols to ''; they now stay NULL so gaps are detectable, and token_pair correspondingly stays NULL (matching enrich_dex_aggregator_trades behavior).

Behavior change note for reviewers: previously-'' symbols become NULL. Anyone filtering token_sold_symbol != '' on 1inch rows sees a difference; this aligns 1inch with how every other dex_aggregator project represents unknown symbols.

Validation: bnb model compiled locally and rendered SQL inspected (fallback chains and tokens join render correctly); relying on CI for the full per-chain compile and tests. Post-merge check: re-run https://dune.com/queries/7709025 — the AR v6 daily blank rate should drop from ~10% to ~0, and the anchor tx's spell row should carry full token details.

[cursor]

PR Summary

Medium Risk
Changes how 1inch AR trade token fields are populated downstream in dex aggregator data; transfer precedence is preserved but NULL vs empty-string semantics shift for consumers.

Overview
Fixes 1inch AR v6 trace-decoded trades (evt_index = -2) that showed empty symbols and NULL amounts when transfer matching failed (e.g. internal router calls from wallet proxies or v4-style vault settlements).

oneinch_ar_executions_macro.sql extends the existing coalesce chains so calldata-decoded token addresses, amounts, and ERC20 metadata are used only after transfer-based resolution. Native sentinels (0x0…0 / 0xeee…e) map to wrapped native + chain symbol via a new tokens lookup join.

oneinch_ar_trades.sql stops forcing unresolved symbols to ''; they stay NULL so quality checks and token_pair behave like other dex_aggregator enrich paths (NULL pair when symbols are unknown).

Reviewer note: rows that previously had '' symbols now have NULL—filters like token_sold_symbol != '' on 1inch AR rows will differ.

^{Reviewed by Cursor Bugbot for commit 7cd9791. Configure here.}