TrinoResult generator cannot recover from errors during lazy segment iteration

[hashhar]

Python generators cannot be resumed after an unhandled exception, the frame is finalized and all subsequent next() calls raise StopIteration.

After #597 spooled segment downloads happen lazily inside the generator during iteration. If a transient error (e.g. S3 timeout) occurs mid-iteration, the caller sees the exception, but the generator is permanently dead.
Subsequent fetchone() calls also return None as if the query completed normally and silently drops remaining rows.

Before #597 this was theoretical issue since fetch() materialized all segments eagerly. I/O errors happened outside the generator.

From looking at other projects it seems the pattern is to use a class with __next__ and __iter__ instead of a generator and having a field on the class to hold state of the iterator so that we can signal to the caller what happened.

Here's a minimal repro:

from trino.client import TrinoResult

class FailingIterator:
    def __init__(self):
        self._count = 0
    def __iter__(self):
        return self
    def __next__(self):
        if self._count >= 3:
            raise IOError("S3 segment download failed")
        self._count += 1
        return [self._count]

class FakeQuery:
    finished = True
    def fetch(self):
        return []

result = TrinoResult(FakeQuery(), FailingIterator())
it = iter(result)

next(it)  # [1]
next(it)  # [2]
next(it)  # [3]
next(it)  # raises IOError, caller sees this and retries
next(it)  # raises StopIteration, caller gets None from fetchone() and thinks query is done