Fix retry counts in cached client

[konstin]

Previously, we dropped the counts from the middleware layer, potentially doing to many retries and/or reporting too few.

Not pretty but fixes the bug.

[konstin]

Removed the optional as zero retries is indistinguishable from None.

[EliteTK]

I'm a little bit too unfamiliar with the underlying code to understand why you consider this to be a hack but I've given it a thorough looking over and I can't see anything that stands out.

The surrounding code may benefit from some future rework, but I haven't looked deeply into it here as it's not really related to this specific PR.

[EliteTK]

nitpick: I think the second comment doesn't make much sense now the addition (consideration) has moved.

Suggested change

	Err(err) => {
	// Check if the middleware already performed retries
	total_retries += err.retries().unwrap_or_default();
	if is_transient_network_error(err.error()) {
	// If middleware already retried, consider that in our retry budget
	let retry_decision = retry_policy.should_retry(start_time, total_retries);
	Err(err) => {
	// Check if the middleware already performed retries and consider it in our retry budget
	total_retries += err.retries().unwrap_or_default();
	if is_transient_network_error(err.error()) {
	let retry_decision = retry_policy.should_retry(start_time, total_retries);

[konstin]

Both of those should be better in the upstack PR: https://github.com/astral-sh/uv/pull/17105/changes#diff-554df8b1efa8f690930e0d00e1c7302751a6e9193fd812a9b1f7305859ceee29R1171-R1201

[EliteTK]

Will look at it tomorrow.

[EliteTK]

Suggested change

	return if total_retries > 0 {
	Err(err.with_retries(total_retries))
	} else {
	Err(err)
	};
	return Err(err.with_retries(total_retries));

Since the retry count is non-optional, conditionally doing with_retries here is now somewhat meaningless.