[bfrancojr]

Fixes #122

Stacked on #123 (the XML converters), whose ProtoToXML carried the first copy of this check. Merge #121, then #123, then this.

sbe::Codec::New now diagnoses a descriptor compiled against the pre-v1.12.0 sbe/annotations.proto the way protowire-go does:

• A file whose (sbe.schema_id) sits at the retired number 50100 gets sbe: file <name> carries (sbe.schema_id) at retired option number 50100 (registered as 1319 since protowire v1.12.0): the descriptor predates ... must be recompiled ....
• A message whose (sbe.template_id) sits at 50200 gets the same diagnosis. It used to be skipped silently, as if it weren't a template, so the codec just didn't know the message.
• A file with no schema id now says sbe: file <name> missing (sbe.schema_id) option, Go's text, where it said <name> missing (sbe.schema_id) file option.

The check, StaleFileError and StaleMessageError in annotations.cc, is now shared with ProtoToXML. A retired number that the descriptor's pool declares as an extension counts as a known option, not a stale one. That's how Go's compiler stores it, and the #123 differential fixture stale.proto pins it.

Evidence

• Tests: SbeStale.CodecNewDiagnosesStaleDescriptors builds descriptors whose options carry 50100 or 50200 as unknown fields, and checks Go's text for both, plus that a plain message is still skipped. SbeStale.CodecNewMissingSchemaIDText checks the new missing text with a pool that declares 50100.
• Mutations, one per check: without the file check, without the message check, and without the pool-declared exemption, a test fails each time.
• Full suite: 655 of 655 pass, 0 error: lines. The --sbe-xml differential output is unchanged and still identical to Go main's. Formatted with clang-format 18.1.8.

[bfrancojr]

Rebased onto #123's new head 212cd45 (this PR's head is now f75124d). The only conflict was CHANGELOG.md, where I kept both entries. Evidence for the pair is on #123: 676 of 676 tests pass, and the differential is clean against Go c13c69b. Against Go main the differential is red because of #134 and #129, not this PR. #132 and #134 will stack on this branch, since they change Codec::New too.

[bfrancojr]

Rebased onto main e1c1b72 (#136 merged); head is now 801d607. The PR's own diff is unchanged, checked by comparing its added and removed lines before and after the rebase. Rebased as part of the SBE stack; evidence is on #123.

[bfrancojr]

Rebased onto main df6b102 (#123 merged); head is now 76f7a4e. The rest of the stack moved with it in one rebase: #137 bcedc19, #140 43eabbe, #143 456dd6e, #147 db34ed3. Each layer's diff against its base is unchanged. Local ctest at the top (#147): 705 of 705. Unmodified scripts/go_differential.sh at #137's head bcedc19, against protowire-go main a1e79dd and spec main 7bde7a9: exit 0, every section ran (954 damaged streams decoded alike, 18 SBE schemas converted alike, 362 documents), no differences. So main's differential should go green once #124 and #137 are in.