A weird log line from the asyncio default exception handler: "Resetting connection with an active transaction"

Sometimes our production component logs the following error
21:05:49.897964000 [Error ] [asyncio] Resetting connection with an active transaction <asyncpg.connection.Connection object at 0x7fb1502561f8>

We create a connection pool first.
Then we use async_timeout, prepared statements, explicit transaction and a cursor to get data.
A coroutine that has this logic may be cancelled sometimes, and this is when we may see the error above. However not every
cancellation produces the error. Cancellation happens oftens but we see this error extremely rarely. I wasn't able to build a
minimal example to reproduce it.

We use a connection pool.
Later, in a different coroutine, we use async_timeout, prepared statements, explicit transaction and a cursor.
A coroutine that has this logic may be cancelled sometimes, and this is when we may see the error above. However not every
cancellation produces the error. Cancellation happens oftens but we see this error extremely rarely. I wasn't able to build a
minimal example to reproduce it.

`
pool = await asyncpg.create_pool(...)

        async with async_timeout.timeout(global_delay_sec) as cm:
            async with pool.acquire() as conn:
                stmt = await conn.prepare(sql)
                columns = stmt.get_attributes()
                ...
                async with conn.transaction():
                    cursor = await stmt.cursor()
                    raw_values = await cursor.fetch(max_rows + 1)
                    if len(raw_values) == max_rows + 1:
                        raise Exception(f"A total number of rows exceeded the allowed maximum of {max_rows} rows")

`

• asyncpg version:
  0.20.1

• PostgreSQL version:
  Reproduced with both postgres-11 and 12

• Do you use a PostgreSQL SaaS? If so, which? Can you reproduce
  the issue with a local PostgreSQL install?:
  No we don't use PostgreSQL SaaS
  I wasn't able to reproce it the issue locally.

• Python version:
  3.6.3

• Platform:
  lsb_release -d
  Description: CentOS Linux release 7.6.1810 (Core)

• Do you use pgbouncer?:
  no

• Did you install asyncpg with pip?:
  yes

• If you built asyncpg locally, which version of Cython did you use?:
  We do not build it locally

• Can the issue be reproduced under both asyncio and
  uvloop?:
  I do use uvloop, I couldn't reproduce it locally. It seems to be an extremely rare issue, happens in our production only once in a month

Start with the reported pool.acquire, prepared-statement, explicit transaction, cursor, and async_timeout flow, focusing on cancellation during the shown operations. Compare behavior under asyncio and uvloop with the PostgreSQL versions listed, and use the production-only log as the symptom to explain. Done means reproducing the rare reset or establishing a confirmed cause and resolution.