non-fast bulk export writes to many files instead of honoring settings

[lmsurpre]

Describe the bug
I have an bulkdata config with a ibm-cos storageProvider and COS settings as follows:

            "cos" : {
              "partUploadTriggerSizeMB": 10,
              "objectSizeThresholdMB": 200,
              "objectResourceCountThreshold": 200000
            },

The fast export seems to be working as desired, but when I add a typeFilter it falls back to the search-based export implementation and the result is that the resources are being split across many more COS objects than they should be.

I turned up tracing and it looks like the currentUploadSize is being computed incorrectly:

2021-08-05 16:47:32.530 wh-cmsiop-cthon-fhir-internal-799fc96897-65rg7 fhir-internal FINE isReadyToCheckpoint [readyToWrite=false (37343 >= 10485760), overSizeThreshold=false (37343 >= 209715200), overResourceCountThreshold=false (100 >= 200000), end=false (1 >= 328)]
2021-08-05 16:47:32.674 wh-cmsiop-cthon-fhir-internal-799fc96897-65rg7 fhir-internal FINE isReadyToCheckpoint [readyToWrite=false (73858 >= 10485760), overSizeThreshold=false (111201 >= 209715200), overResourceCountThreshold=false (200 >= 200000), end=false (2 >= 328)]
...
2021-08-05 16:47:44.294 wh-cmsiop-cthon-fhir-internal-799fc96897-65rg7 fhir-internal FINE isReadyToCheckpoint [readyToWrite=false (3796495 >= 10485760), overSizeThreshold=false (199361860 >= 209715200), overResourceCountThreshold=false (10400 >= 200000), end=false (104 >= 328)]
2021-08-05 16:47:45.738 wh-cmsiop-cthon-fhir-internal-799fc96897-65rg7 fhir-internal INFO pushFhirJsonsToCos: '3905969' bytes were successfully appended to COS object - '0vzo_7VIZWnkThMov9MmU8AO7JdwylM9jHTEEGd5JD0/Patient_1.ndjson' uploadId='0100017b-1849-57a1-aa48-8cc67e858702'

Specifically, note how the "sizeThreshold" bytes are growing faster than the "readyToWrite" bytes even before we've started an upload and cleared the buffer.

Environment
Which version of IBM FHIR Server?

To Reproduce
Steps to reproduce the behavior:

1. configure the server with a bulkdata storageProvider that uses s3 (e.g. aws-s3 or ibm-cos)
2. issue an export command with a typeFilter (e.g. GET [base]/$export?_type=Patient&_typeFilter=Patient?_elements=gender
3. note that the results have more COS objects than it should

Expected behavior
The system should continue writing to a single cos object until either the objectSizeThresholdMB or the objectResourceCountThreshold has been reached

Additional context

[kmbarton423]

Ran a variety of $export scenarios (type=file) against 250K resources focusing on system level and patient. Also used _type and _typeFilter query parm combos in a subset of the $export(s). Varied config parms and confirm honored ( ie writeTriggerSizeMB, sizeThresholdMB, resourceCountThreshold).

Ran bulk data sniff test which utilizes type=ibm-cos.