Skip namespace check for InMemoryKVS

[gsheasby]

Goals (and why):
We don't care about the namespace for InMemoryKVS, so we want to allow
any namespace to be set. Previously, each product that used InMemoryKVS
for their tests needed to configure the namespace to be exactly "test".

Implementation Description (bullets):

• Set the default namespace to empty for InMemoryKVS
• Ignore part of the config check when the KVS config is an InMemoryAtlasDbConfig

Concerns (what feedback would you like?): Is the config check ignoring too hacky?

Where should we start reviewing?: wherever, it's pretty small

Priority (whenever / two weeks / yesterday): this week, preferably this release as it's been hurting internal testing efforts.

---

This change is 

[fsamuel-bs]

This change allows products to start without both a keyspace and a namespace config on InMemoryKVS. If none of the configs are specified we seem to fail on startup if the timelock block is present — https://github.com/palantir/atlasdb/blob/develop/atlasdb-config/src/main/java/com/palantir/atlasdb/factory/TransactionManagers.java#L588. We should probably make this a test case in TransactionManagersTest, as I don't know the behavior of timelock with client being empty string.

With this addition, I think this PR should be ready to merge.

[codecov-io]

Codecov Report

Merging #2541 into develop will increase coverage by 0.03%.
The diff coverage is 100%.

@@              Coverage Diff              @@
##             develop    #2541      +/-   ##
=============================================
+ Coverage      60.03%   60.07%   +0.03%     
+ Complexity      3389     2466     -923     
=============================================
  Files            860      859       -1     
  Lines          40119    40079      -40     
  Branches        4070     4073       +3     
=============================================
- Hits           24084    24076       -8     
+ Misses         14569    14540      -29     
+ Partials        1466     1463       -3

Impacted Files	Coverage Δ	Complexity Δ
...palantir/atlasdb/memory/InMemoryAtlasDbConfig.java	83.33% <100%> (ø)	5 <1> (+5)	⬆️
...ava/com/palantir/atlasdb/config/AtlasDbConfig.java	98% <100%> (-2%)	0 <0> (-24)
...antir/nexus/db/pool/HikariCPConnectionManager.java	52.75% <0%> (-3.15%)	14% <0%> (-2%)
...main/java/com/palantir/paxos/PaxosLearnerImpl.java	87.8% <0%> (-2.44%)	0% <0%> (ø)
.../com/palantir/atlasdb/table/common/TableTasks.java	57.69% <0%> (-0.35%)	20% <0%> (+20%)
...tlasdb/table/description/render/TableRenderer.java	68.41% <0%> (-0.21%)	0% <0%> (-9%)
...asdb/transaction/impl/AbstractTransactionTest.java	96.5% <0%> (-0.01%)	103% <0%> (+1%)
...alantir/atlasdb/memory/InMemoryAtlasDbFactory.java	21.21% <0%> (ø)	5% <0%> (+5%)	⬆️
...asdb/transaction/impl/SerializableTransaction.java	88.41% <0%> (ø)	86% <0%> (+86%)	⬆️
... and 27 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 59782d4...a8a0f30. Read the comment docs.

[fsamuel-bs]

I think we could rewrite this as:

if (kvs is InMemory) {
    (namespacePresent && namespaceEqualsClientAndKeyspace) || timelockIsNotEmpty
} else {
    (namespacePresent && namespaceEqualsClientAndKeyspace) || (keyspacePresent && keyspaceEqualsClient)
}

Feel free to improve it even more. My goal here was to make it as descriptive as possible.

[fsamuel-bs]

I'm going to approve it as is now, as we could refactor it later. The tests prove that we have the behavior we want here.