feat(io): restore the oss scheme with a round trip test and docs

[plusplusjiajia]

Per the suggestion on #889: oss comes back with a test and documentation.

oss returns to kS3Schemes, which also restores the oss:// credential prefix now that IsS3CredentialPrefix derives from that list. A bare oss prefix stays rejected, so nothing widens beyond what main had before #889. CanonicalizeS3Scheme now derives from the same list too — an alias missing there would not
fail, it would silently stop matching its credential and fall back to the default one.

rest_arrow_file_io_test.cc and its CMake target are restored; #889 removed them, leaving REST -> ResolvingFileIO -> registry -> Arrow FileIO uncovered. The new test writes, reads back and deletes an object addressed as oss://, against ICEBERG_TEST_S3_URI — the MinIO the AWS jobs already start — and skips when
unset. It passes in 99 ms locally and fails with URI scheme 'oss' is not supported for FileIO resolution once oss leaves the scheme table. It proves routing, not credential matching: the same keys also reach the default client through the AWS chain.

file-io.md gains an S3 property table and a section on S3-compatible storage. The OSS example is verified, not transcribed: the signing region is cn-hangzhou while the endpoint carries the oss- prefix, and path-style is rejected with SecondLevelDomainForbidden: Please use virtual hosted style to access.

[wgtmac]

How about still removing oss support from #889 and then you add it back by adding test and documentation to fileio.md to instruct users how to config for oss?

[plusplusjiajia]

How about still removing oss support from #889 and then you add it back by adding test and documentation to fileio.md to instruct users how to config for oss?

@wgtmac Works for me — I'll keep this PR and grow it into that: once #889 lands I'll rebase, restore the scheme, and add fileio.md covering how to configure it.

[wgtmac]

#889 has been merged. You may want to proceed with a rebase.

[plusplusjiajia]

#889 has been merged. You may want to proceed with a rebase.

@wgtmac Done — rebased and grown into the restore we discussed: oss back in kS3Schemes, the round-trip test restored, and file-io.md extended with S3 configuration and an OSS section. The test also caught a Windows regression on its first run — ParseScheme took drive letters for URI schemes — fixed minimally inlocation_util.cc

[wgtmac]

I just took an initial review. Have you checked it against a real OSS setup to make sure it works?

[plusplusjiajia]

I just took an initial review. Have you checked it against a real OSS setup to make sure it works?

@wgtmac Thanks for the review! Yes — verified against a real OSS bucket (that's where the addressing/endpoint guidance in file-io.md comes from).

[wgtmac]

LGTM. Thanks @plusplusjiajia for working on this!