Wrong Aave V3 PoolDataProvider address in default.toml + fork.toml

[obchain]

Symptom

Bot scans correctly post-#418, identifies all 4 demo borrowers as liquidatable on the BSC fork, but every flash-loan quote fails:

WARN charon_flashloan::router: source skipped: quote failed
  source=AaveV3
  err=Rpc("getReserveConfigurationData: ABI decoding failed: buffer overrun while deserializing")
WARN charon_flashloan::router: no flash-loan source could cover the borrow

Result: liquidatable=4 queued=0. Bot detects the opportunities but never builds calldata or broadcasts.

Root cause

The Aave V3 BSC PoolDataProvider address baked into both config/default.toml:78 and config/fork.toml:77 is 0x41393e5e337606dc3821075Af65AeE84D7688CBD. There is no contract at that address on BSC mainnet — cast code against the live BSC chain and against any forked block returns 0x. Calls go to an EOA / empty slot, get back zero bytes, and alloy reports buffer overrun because it expected a tuple worth of data.

The correct address can be resolved on-chain at any time via the configured addresses_provider:

$ cast call 0xff75B6da14FfbbfD355Daf7a2731456b3562Ba6D 'getPoolDataProvider()(address)' \
    --rpc-url http://127.0.0.1:8545 --block 91323624
0xc90Df74A7c16245c5F5C5870327Ceb38Fe5d5328

The Aave V3 Pool (0x6807dc923806fE8Fd134338EABCA509979a7e0cB) was correct — only the data provider drifted.

Proposed fix

Update both configs:

data_provider = "0xc90Df74A7c16245c5F5C5870327Ceb38Fe5d5328"

Add a startup probe (AaveFlashLoan::connect) that resolves addresses_provider.getPoolDataProvider() and warns if the configured data_provider does not match. Catches address drift before the first quote silently fails.

Impact

Hard blocker on every fork demo and every mainnet run alike — flash-loan routing has no working source, so nothing is ever broadcast. Surfaced while validating #412/#413/#414 + #418 end-to-end at the Notion demo block 91323624.