Assorted sync metadata storage refactoring

[tgoyne]

This is pulling some of the incidental changes out of #7300 since that PR is growing far too large. None of this has meaningful functional changes on its own and is mostly just pushing code around.

One of the upcoming changes is making SyncUsers lazily loaded and reloaded when the data in the metadata store changes, which benefits from it being relatively cheap to validate that the stored user is valid. Changing RealmJWT to operate on a string view had cascading effects on base64 and exjson parsing, which I also updated to use Span rather than a pointer+size while I was touching it.

Checking if the data in the metadata realm is valid was done incompletely in three different places, and now is just in SyncMetadataManager. File action handling is also there now as that's going to be required for it to be multiprocess safe. Most of SyncMetadataManager's API is now unused outside of tests, but is left in place for now as removing it requires rewriting the tests (which is done in #7300).

CheckedMutex::lock() is const, so it doesn't have to be declared as mutable.

[coveralls-official]

Pull Request Test Coverage Report for Build thomas.goyne_158

• 219 of 246 (89.02%) changed or added relevant lines in 23 files are covered.
• 82 unchanged lines in 9 files lost coverage.
• Overall coverage increased (+0.005%) to 91.863%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/realm/object-store/c_api/app.cpp	0	1	0.0%
src/realm/sync/noinst/protocol_codec.cpp	0	1	0.0%
src/realm/sync/noinst/server/access_token.cpp	1	3	33.33%
src/realm/object-store/sync/impl/sync_metadata.cpp	94	97	96.91%
src/realm/util/file.cpp	3	6	50.0%
src/realm/object-store/sync/jwt.cpp	49	57	85.96%
src/realm/object-store/sync/jwt.hpp	3	12	25.0%

Files with Coverage Reduction	New Missed Lines	%
src/realm/index_string.cpp	1	87.85%
src/realm/query_expression.hpp	2	93.51%
src/realm/sync/network/websocket.cpp	2	74.44%
src/realm/util/serializer.cpp	2	90.91%
src/realm/sync/transform.cpp	4	63.01%
test/fuzz_tester.hpp	5	58.95%
src/realm/bplustree.cpp	7	75.72%
src/realm/sync/noinst/server/server.cpp	14	76.62%
test/fuzz_group.cpp	45	51.75%

Totals
Change from base Build 2021:	0.005%
Covered Lines:	235332
Relevant Lines:	256176

---

💛 - Coveralls

[danieltabacaru]

should we delete all users in a single transaction? The downside would be that in case of an exception (other than FileAccessError), no user is removed.

[tgoyne]

I am currently undecided on what level of batching is correct. This PR is matching the existing behavior and it's just much more obvious that we're performing write transactions in a loop due to the code movement.

[danieltabacaru]

If I understand these changes correctly, we got rid of AppMetadata table right?

[tgoyne]

The AppMetadata table won't be deleted from existing files (that requires a schema version bump and a migration), but it won't be created in newly initialized metadata realms.

[danieltabacaru]

The AppMetadata table won't be deleted from existing files (that requires a schema version bump and a migration), but it won't be created in newly initialized metadata realms.

Yes, that's what I meant.