enhancement: EVM native balance deltas require trace API support

[user1303836]

Summary

EVM `NativeBalanceDelta` extraction is intentionally unimplemented (documented in `adapters/src/evm_parser.rs:513-530`). The current workaround infers ETH transfers from the transaction-level `value` field, which is unreliable:

1. Failed transactions: A tx with `value=1 ETH` that reverts is still recorded as transferred
2. Internal transfers: Contract-to-contract ETH transfers are invisible without traces
3. Wrapped ETH: Sending ETH to WETH contract is recorded as a transfer, not a mint
4. Gas refunds and self-destructs: Not captured

Impact

• No accurate ETH balance change tracking for any EVM chain
• Tax software cannot compute correct ETH positions
• Internal contract interactions (flashloans, DEX routing) are invisible

Recommended Approach

1. Implement Bronze-layer trace ingestion (`debug_traceTransaction` or `trace_transaction`)
2. Create `raw_evm_traces` table
3. Extract pre/post balances from trace data for NativeBalanceDelta materialization
4. Add E2E tests against mainnet fork with complex contract interactions

🤖 Generated with Claude Code

[user1303836]

After investigating the current EVM parser code and Bronze layer storage, this is clearly too large for a single PR. Here's a scoping breakdown:

What's needed end-to-end:

1. A new Bronze-layer trace ingestion path (debug_traceTransaction or trace_transaction) — this requires RPC provider support (not all nodes expose debug/trace APIs)
2. A new raw_evm_traces table and migration
3. A parser that extracts pre/post balance state diffs from trace data
4. Integration with the existing NativeBalanceDelta placeholder in evm_parser.rs (line 544-549)
5. E2E tests against realistic trace data (mainnet fork or fixtures)

Recommended phased approach:

• Phase 1: Add the raw_evm_traces migration and a TraceIngestor trait stub. This creates the schema without any runtime behavior change. (~1 PR)
• Phase 2: Implement trace fetching for a single provider (e.g., debug_traceTransaction with Geth-style callTracer). Store raw traces in Bronze. (~1 PR)
• Phase 3: Parse traces into NativeBalanceDelta ledger entries, replacing the current value-field heuristic. Add integration tests with trace fixtures. (~1 PR)

Each phase is independently mergeable and testable. Phase 1 alone is not useful without phases 2-3, so it may make sense to combine phases 1 and 2 as the first deliverable.

Skipping implementation for now — this needs to be planned as a multi-PR effort.