multitenant: don't panic if reader doesn't exist yet

[stevendanna]

While shared-process tenant servers are not likely to make requests
before the capability reader exists, the limiter factory looks up the
relevant tenant ID from the start key of the range descriptor and it
isn't unlikely that we'll see requests against tenant ranges before we
have a capability reader available.

Epic: none

Release note: None

Release justification: Low risk fix to avoid panics in tests.

[blathers-crl]

It looks like your PR touches production code but doesn't add or edit any test code. Did you consider adding tests to your PR?

_{🦉 Hoot! I am a Blathers, a bot for CockroachDB. My owner is dev-inf.}

[cockroach-teamcity]

This change is 

[ajstorm]

LGTM

[ecwall]

This looks fine, but I am worried that these functions are diverging. They are all pretty much just checking a bool(s), but have slight differences already.

[arulajmani]

Should we add a comment inline as well, in case future readers wonder why this function doesn't panic like the others?

[knz]

Maybe update commit message and PR description to replace "in memory tenants" by "shared process tenant servers".

[stevendanna]

Comments and commit message updated. Will merge once CI is green.

@ecwall

This looks fine, but I am worried that these functions are diverging. They are all pretty much just checking a bool(s), but have slight differences already.

I kinda wonder if what we want here is a HasCapability(ctx, tenID, cap) and HasCapability(ctx, tenID, cap, authorizer.ReturnDefaultIfUnavailable()). Not going to do that in this PR, but happy to chat about how this should evolve.

[stevendanna]

Test failure is slow quiesce.

bors r=knz,ajstorm,arulajmani

[craig]

Build succeeded:

• Bazel Essential CI (Cockroach)