Ensure re-broadcasted mempool txs are "unpruned"

[zone117x]

Mempool transactions need to be "restored" in the following scenario:

1. Mempool tx is received
2. Mempool tx is dropped after not being mined for a period of time
3. Mempool tx is re-broadcasted

It seems like step 3 may not be handled properly during re-ingestion of the mempool tx event:

stacks-blockchain-api/src/datastore/pg-write-store.ts

Lines 1770 to 1771 in c4cba69

	INSERT INTO mempool_txs ${sql(values)}
	ON CONFLICT ON CONSTRAINT unique_tx_id DO NOTHING

The ON CONFLICT might need to instead 1) ensure the tx isn't already mined because these events can come out-of-order from /new_block, then 2) update pruned to false.

cc @lgalabru who identified this potential bug

[zone117x]

Building off the query which is being updated in #1815, I think the fix would look something like:

WITH settled AS (
    SELECT tx_id
    FROM txs
    WHERE
        tx_id IN ${sql(values.map(b => b.tx_id))} AND
        canonical = true AND
        microblock_canonical = true
),
inserted AS (
    INSERT INTO mempool_txs ${sql(values)}
    ON CONFLICT (tx_id) DO 
    UPDATE SET
        pruned = CASE 
            WHEN EXCLUDED.tx_id NOT IN (SELECT tx_id FROM settled) THEN false
            ELSE mempool_txs.pruned
        END
    RETURNING tx_id
),
pruned AS (
    UPDATE mempool_txs
    SET pruned = true
    WHERE
        tx_id IN (SELECT tx_id FROM settled) AND
        pruned = false
    RETURNING tx_id
),
count_update AS (
    UPDATE chain_tip SET
        mempool_tx_count = mempool_tx_count + (SELECT COUNT(*) FROM inserted),
        mempool_tx_count = mempool_tx_count - (SELECT COUNT(*) FROM pruned),
        mempool_updated_at = NOW()
)
SELECT tx_id FROM inserted

[zone117x]

Closed by #1823