Resource attribute decoding corrupts unencoded non-ASCII characters #8890

Describe the bug
When an OTEL_RESOURCE_ATTRIBUTES value contains both a percent-encoded sequence and an unencoded non-ASCII character, the non-ASCII character is corrupted.

Steps to reproduce
Set OTEL_RESOURCE_ATTRIBUTES=key=café%20bar (or -Dotel.resource.attributes=key=café%20bar) and build the SDK with autoconfigure.

What did you expect to see?
key="café bar". The spec only requires , and = to be percent-encoded, so an unencoded é is valid input. Before #8059, URLDecoder kept such characters.

What did you see instead?
key="caf� bar". EnvironmentResource.decodeResourceAttributes() writes each unencoded char into a byte array with (byte) c and decodes the array as UTF-8, so any char at or above U+0080 is truncated. Values without % skip decoding and are not affected.

What version and what artifacts are you using?
Artifacts: opentelemetry-sdk-extension-autoconfigure (also declarative config resource.attributes_list, which uses a copy of this class)
Version: 1.60.0 through 1.66.0, main @ 7f423cb
How did you reference these artifacts? N/A

Environment
Compiler: Temurin 21
OS: N/A

Additional context
Regression from #8059, which replaced URLDecoder to keep + as a literal (#8030).
Spec: https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/resource/sdk.md#specifying-resource-information-via-an-environment-variable