More granular read timeout

We are currently affected by cache read timeouts. The pattern is similar to https://github.com/actions/cache/issues/1141: cache download starts normally and then download stops completely.

There is a SEGMENT_DOWNLOAD_TIMEOUT_MINS environment variable but it only affects a full segment (1 or 2 GB).

Would it be possible to have more granular control over the timeout? Maybe something like:

• SEGMENT_CONNECT_TIMEOUT_SECS: Timeout if it takes more than X seconds to establish the connection to the remote cache.
• SEGMENT_REAT_TIMEOUT_SECS: Timeout if there is more than X seconds without a single byte being read on the wire.
• SEGMENT_DOWNLOAD_TIMEOUT_MINS: keep current behaviour. Timeout if it takes more than X minutes to download a segment.

This way, we don't have to wait 10min if a relatively small entry times out

Start by reviewing the existing SEGMENT_DOWNLOAD_TIMEOUT_MINS behavior and the cache segment restore timeout guidance in tips-and-workarounds.md, then compare the failure pattern with actions/cache#1141. Define how connection, idle-read, and full-segment timeouts should interact, and verify that each proposed environment variable has documented behavior and coverage.