Update synchronizer.Sync2 to Sync from legacy impl

Reorg l2db before starting pipeline

Set gin debug mode via config

Calculate ForgeBatch gasLimit with parametrized formula

Fix missing timer reset in TxManager

Add API metric for time to forge L1 tx

Fix ZKI Exit Balance2 accumulate Amounts

Add new config setting `Debug.GinDebugMode`.  When set to true, gin will run in
debug mode.  If not set, gin will run in release mode.  Before this change, gin
always ran in debug mode, so to keep the same behaviour as before, set this
parameter to true

Add the following config parameters at
`Coordinator.EthClient.ForgeBatchGasCost`:
- `Fixed`
- `L1UserTx`
- `L1CoordTx`
- `L2Tx`
Which are the costs associated to a ForgeBatch transaction, split
into different parts to be used in the formula to compute the gasLimit.

Also, replace usage of time.Duration by time.NewTimer, because the later allows
replacing timers by stopping them before so that we never leak resources.

In http servers, first listen, then serve

Wait for serverProof before starting batch

Change how TX per Batch is calculated in state API endpoint

Diferentiate SQL connections by read write

Update ZKI generation to match circuit behaviour for Exit Amount>Balance

Currently the circuit does not use an Exit Leaf at the Exit MerkleTree
when the Amount=0 & EffectiveAmount=0, but it does use an Exit Leaf when
the Amount>0 but EffectiveAmount=0 (for example when tx.Amount >
sender.Balance). This is a particularity of the approach of the circuit,
the idea will be in the future to update the circuit and when Amount>0
but EffectiveAmount=0, to not add the Exit in the Exits MerkleTree, but
for the moment the Go code is adapted to the circuit and when an Exit
with Amount>0 & EffectiveAmount=0, it will be added to the Leaf of the
Exit MerkleTree.

- In test, doing the `net.Listen` outside of the goroutine guarantees that we
  are listening, so we avoid a possible datarace consisting of doing an http
  request before listening.
- In packages that run an http server: doing the listen first allows
    - Checking for errors when opening the address for listening before
      starting the goroutine, so that if there's an error on listen we can
      terminate grafecully
    - Logging that we are listening when we are really listening, and not
      before.

Add Prometheus initial setup

In API recommended fee, use minFeeUSD as min value

fix possible runtime error

node: remove redundant error check

tx manager: remove redundant error check

Stop using stateDB in API

Add minPriceUSD in L2DB, check maxTxs atomically

- Add config parameter `Coordinator.L2DB.MinPriceUSD` which allows rejecting
  txs to the pool that have a fee lower than the minimum.
- In pool tx insertion, checking the number of pending txs atomically with the
  insertion to avoid data races leading to more than MaxTxs pending txs in the
  pool.

Delete pending txs by external mark, store tx IP

- In tx_pool, add a column called `external_delete` that can be set to true
  externally.  Regularly, the coordinator will delete all pending txs with this
  column set to true.  The interval for this action is set via the new config
  parameter `Coordinator.PurgeByExtDelInterval`.
- In tx_pool, add a column for the client ip that sent the transaction.  The
  api fills this value using the ClientIP method from gin.Context, which should
  work even under a reverse-proxy.

Feature/accountupdatetable

Txs w/ exit Amount=0,to not create new Exit leafs

There are 2 ways to interpret an Exit transaction with Amount=0:
`A`. Not adding a new Leaf to the ExitTree, so not modifying the Exit
MerkleRoot
`B`. Adding a new Leaf to the ExitTree, with Balance=0, which modifies
the Exit MerkleRoot

Currently the [Circuits](https://github.com/hermeznetwork/circuits) are
doing approach `A`, and the
[hermez-node](https://github.com/hermeznetwork/hermez-node) is doing the
approach `B`. The idea of this commit, is to use approach `A` also in
the hermez-node.

Duplicated error when caused by meddler