Refactor[mqbs::FileStore]: Encapsulate write-head cursor accessors

[kaikulimu]

Summary

• Consolidates access to the FileStore "write head" — the (leaseId, seqNum) cursor identifying the next record this store writes or applies — behind a small, intention-revealing accessor family, instead of scattered raw field/map manipulation.
• Follows up the earlier write-head renames (member d_primaryLeaseId → d_writeHeadLeaseId in Refactor[mqbs::FileStore]: Rename d_primaryLeaseId to d_writeHeadLeaseId #1660, accessor primaryLeaseId() → writeHeadLeaseId() in Refactor[mqbs::DataStore]: Rename primaryLeaseId to writeHeadLeaseId #1666): the two halves of the cursor now have symmetric names and the leaky mutable-reference accessor is gone.
• Pure refactoring — no behavior change.

[678098]

Should this method has any preconditions such as d_writeHeadLeaseId <= leaseId?

[kaikulimu]

Not yet, in legacy mode in FileStore::clearPrimary() there is a write head rollbacks mechanism. I am trying to eliminate rollbacks but will do it across a series of PRs to make the changes gradual.

[678098]

.at might throw an exception and the current design choice is not to throw exceptions in the code:

    at(const LOOKUP_KEY& key) {
        HashTableLink *node = d_impl.find(key);

        if (!node) {
            BloombergLP::bslstl::StdExceptUtil::throwOutOfRange(
                      "unordered_map<...>::at(LOOKUP_KEY): invalid key value");
        }
        return static_cast<HashTableNode *>(node)->value().second;
    }

What about changing it to

auto iter = d_highestSeqNums.find(d_writeHeadLeaseId);
BSLS_ASSRT_SAFE(iter != d_highestSeqNums.end());
++iter.second;

[kaikulimu]

No auto in C++03, but I will do the assert instead.

[678098]

Found it weird that DataStoreRecordKey has inverted order of args compared to basically everything in this file, for example:

            setWriteHead(recHeader->primaryLeaseId(),
                         recHeader->sequenceNumber());

[kaikulimu]

I can also fix it in this PR. Still pure refactoring so won't be too much.

[678098]

Now member alignment is not ideal:

d_primaryLeaseId - 4 bytes
[empty space] - 4 bytes
d_sequenceNum - 8 bytes

If only we could use 64bit type for leaseId too...

[kaikulimu]

Switched them back only in the data member declarations.

[678098]

Suggested change

	const bsls::Types::Uint64 sequenceNum);
	bsls::Types::Uint64 sequenceNum);

This is a basic type, no need to have const

[678098]

There are 2 lookups now when we can have 1 for better performance.

BSLS_ASSERT_SAFE(d_highestSeqNums.find(d_writeHeadLeaseId) !=
                     d_highestSeqNums.end());

This check searches for iterator and immediately drops it, when it can be reused to increment the value

[678098]

Looks good