fix(storage): report AWS credential failures as I/O

[lance-gatefixer]

Root cause

AwsCredentialAdapter classified failures from the external AWS credential provider chain as Lance internal errors. In managed environments such as AWS Glue, a missing or unreachable credential source therefore produced the misleading internal-bug wrapper shown in the issue instead of an actionable I/O/configuration-path failure.

Fix

• Classify AWS credential resolution failures as I/O errors while retaining the detailed provider diagnostic.
• Add a regression test that injects a failed credential provider and verifies the I/O variant, nested cause, and absence of the internal-bug message.

Validation

• cargo fmt --all
• cargo clippy --all --tests --benches -- -D warnings
• cargo test -p lance-io test_aws_credential_failure_is_io_error
• cargo test -p lance-io -- --skip uring::tests (197 unit tests and doctests passed)

A full cargo test -p lance-io was also attempted; unrelated Linux io_uring cases failed or hung in this container, so the passing package run excluded that module.

Fixes #2466

[lance-gatekeeper]

✅ Gate recommendation: approve.

The change corrects the error category at the AWS credential-adapter boundary while preserving the provider diagnostic. The injected-provider regression test covers the propagated object-store error and confirms it no longer presents an external credential failure as an internal Lance defect.