Fix array latest schema selection for same MS timestamps schemas.

[KiterLuc]

PR #4549 attempted to fix the latest array schema selection. Unfortunately, it didn't take into account that some array schemas might have the same timestamps. This change uses the fact that array schema URIs are lexicographically sorted (names generated in the same timestamps use extra sorting bits in the UUID) to chose the latest array URI whilst keeping the original intent of #4549 and checking against the end timestamp.

[sc-49806]

---

TYPE: BUG
DESC: Fix array latest schema selection for same MS timestamps schemas.

[teo-tsirpanis]

Thanks!

[ihnorton]

This seems to indicate that we are missing explicit test coverage for schema evolution within the same timestamp -- please make sure that is added in follow-up.

[eddelbuettel]

Just FYI I recently 'adopted' a CRAN package wrapping ULID (via a small C++ library which actually needed a small patch to work correctly with milliseconds via std::chrono, see here) but once you use that you get nicely sorted identifiers, and they can be unmarshaled back too. Resolution is millies as we do:

> ids <- ulid::ulid(3)
> ids
[1] "01J1S4AWSGXJXSMCPB8XMMBBVP" "01J1S4AWSGS2RHQQRBAZAKK264" "01J1S4AWSGGZQXRH38HV3GBJNB"
> ulid::unmarshal(ids)
                   ts              rnd
1 2024-07-02 02:07:52 XJXSMCPB8XMMBBVP
2 2024-07-02 02:07:52 S2RHQQRBAZAKK264
3 2024-07-02 02:07:52 GZQXRH38HV3GBJNB
> 
> ulid::unmarshal(ids)[,1]     # show milliseconds
[1] "2024-07-02 02:07:52.496 CDT" "2024-07-02 02:07:52.496 CDT"
[3] "2024-07-02 02:07:52.496 CDT"
> 

It is based on a pretty small MIT licensed library if you want to poke.

[KiterLuc]

This seems to indicate that we are missing explicit test coverage for schema evolution within the same timestamp -- please make sure that is added in follow-up.

This fix is coming from tests failing in the nightlies.