Fix: improve snap piece cleanup / selection

[magik6k]

• Add a delete button for failed sectors in the snap pipeline page
• Cleanup piece refs after encode in SnapDeals (this is where we finalize), this was leaving piece files around
• Make snap sector selection if there isn't already active snap for candidate sectors

Last one was causing

2024-08-12T17:24:00.412+0200	WARN	rpc	go-jsonrpc@v0.5.1-0.20240730181539-2bb4fbb48fdd/handler.go:474	error in RPC call to 'Filecoin.SectorAddPieceToAny': allocating sector numbers:
    github.com/filecoin-project/curio/market.(*PieceIngesterSnap).AllocatePieceToSector
        /home/magik6k/curio/market/deal_ingest_snap.go:372
  - adding deal to sector:
    github.com/filecoin-project/curio/market.(*PieceIngesterSnap).AllocatePieceToSector.func1
        /home/magik6k/curio/market/deal_ingest_snap.go:364
  - ERROR: Conflict detected for piece_index 0 (SQLSTATE P0001)

When adding a bunch of sectors at the same time

[LexLuthr]

I see that we were hit by the curse of transactions.