fix(gamification): decode real payment amounts from XDR, classify don…

[Martins0152]

…ation/distribution/claim correctly

Fixes #142

• transaction-indexer.ts no longer uses fee_charged as the donation amount. It now decodes each transaction's envelope_xdr via the Stellar SDK XDR parser to get the real payment amount and asset code.
• Introduces src/lib/horizon/envelope-decode.ts: shared low-level XDR primitives (getEnvelopeOperations, decodePaymentOp, decodeInvokeHostFunctionOp, mapSorobanFunctionName, asset code decoding) extracted from decode-horizon-transaction.ts so both the real-time dashboard stream and the gamification indexer decode XDR the same way instead of duplicating the logic.
• decode-horizon-transaction.ts is refactored to use the shared primitives; its public behavior/output for useRealTimeTransactions is unchanged.
• TransactionEvent.type is now classified from real operation
  semantics: outgoing payments -> donation, incoming payments ->
  distribution, invokeHostFunction calls matching claim/distribute Soroban function names -> claim/distribution. Soroban 'donate'-style calls are intentionally not counted as donations since no verifiable amount can be recovered from the operation without decoding contract arguments (out of scope).
• recipient is now the actual payment destination or contract ID, fixing unique_campaigns_supported.
• Adds resumable cursor support: indexTransactions now honors options.cursor (already persisted via badge-storage.ts's lastSyncedCursor) and issues zero extra Horizon calls when there are no new transactions.
• TransactionEvent/IndexResult/indexTransactions signature, badge-rules.ts, badge-evaluator.ts, and badge-storage.ts are all unchanged per the issue's constraints.

Tests: src/lib/gamification/tests/transaction-indexer.test.ts covers XDR-decoded amounts, claim classification, fee_charged vs real amount, cursor resumption, AID token currency, undecodable XDR handling, and a badge-evaluator integration scenario.

npm run test and npm run type-check pass with no new regressions (pre-existing unrelated failures on master are unaffected).