Ensure that ScanInfo is cached before callbacks fire

Fixes a regression that was introduced in aosp/1312940.

There is a critical section in requestNetworkScan after the scanId
has been generated and before the ScanInfo is added to the cache.
If during this time a callback fires on a separate thread, it will
not yet find the ScanInfo record, and due to strict checking, this
causes a RuntimeException. This CL re-expands the synchronized block
to cover the critical section, and clarifies a little bounds checking.

Note: this change is not ideal. This means that all callbacks for all
scans will be blocked while a request for a new scan is being made,
which was the motivation behind the change that introduced this
regression. But, a narrow performance hit is better than a crash, and
it's not worth the complexity of a bigger fix.

Bug: 200634560
Test: compilation
Change-Id: I4670da109256170121ceb6d8fbad0efda310335f