Fix client reset cycle detection for PBS recovery errors

[tgoyne]

Tracking that a client reset was in progress was done in the same write transaction as the recovery operation, so if recovery failed the tracking was rolled back too. This worked for FLX due to that codepath committing before beginning recovery.

The good news is that we don't have any reports of users running into this problem, which is a strong sign that client reset recovery failing isn't a common scenario in the first place.

[coveralls-official]

Pull Request Test Coverage Report for Build github_pull_request_285034

• 31 of 31 (100.0%) changed or added relevant lines in 3 files are covered.
• 672 unchanged lines in 18 files lost coverage.
• Overall coverage decreased (-0.04%) to 91.677%

---

Files with Coverage Reduction	New Missed Lines	%
src/realm/query_engine.hpp	2	94.17%
src/realm/sync/noinst/protocol_codec.hpp	3	76.72%
src/realm/util/file.cpp	3	82.2%
src/realm/unicode.cpp	4	90.15%
src/realm/util/assert.hpp	4	87.1%
src/realm/alloc_slab.cpp	5	92.89%
src/realm/sync/noinst/client_impl_base.cpp	6	85.25%
test/fuzz_group.cpp	8	53.37%
src/realm/sync/noinst/server/server.cpp	16	76.16%
src/realm/index_string.hpp	22	82.86%

Totals
Change from base Build 1847:	-0.04%
Covered Lines:	231270
Relevant Lines:	252267

---

💛 - Coveralls

[ironage]

Good find!