Reduce MultiprocessRollingFileAppender overhead

[swebb2066]

The benchmark on Ubuntu 22.04 gcc 11.4 are

Benchmark	Time	CPU	Iterations
Testing disabled logging request	0.474 ns	0.474 ns	1000000000
Testing disabled logging request/threads:6	0.139 ns	0.835 ns	819039978
Appending 5 char string using MessageBuffer, pattern: %m%n	342 ns	342 ns	2045248
Appending 5 char string using MessageBuffer, pattern: %m%n/threads:6	418 ns	2350 ns	293568
Appending 49 char string using MessageBuffer, pattern: %m%n	387 ns	387 ns	1803925
Appending 49 char string using MessageBuffer, pattern: %m%n/threads:6	496 ns	2717 ns	229734
Appending int value using MessageBuffer, pattern: %m%n	530 ns	530 ns	1340650
Appending int value using MessageBuffer, pattern: %m%n/threads:6	504 ns	2809 ns	250008
Appending int+float using MessageBuffer, pattern: %m%n	1041 ns	1041 ns	669224
Appending int+float using MessageBuffer, pattern: %m%n/threads:6	625 ns	3597 ns	190068
Appending int value using MessageBuffer, pattern: [%d] %m%n	569 ns	569 ns	1229368
Appending int value using MessageBuffer, pattern: [%d] [%c] [%p] %m%n	629 ns	629 ns	1117287
Appending 49 char string using FMT, pattern: %m%n	358 ns	358 ns	1941068
Appending 49 char string using FMT, pattern: %m%n/threads:6	481 ns	2638 ns	265746
Appending int value using FMT, pattern: %m%n	407 ns	407 ns	1760876
Appending int value using FMT, pattern: %m%n/threads:6	482 ns	2669 ns	256482
Appending int+float using FMT, pattern: %m%n	536 ns	536 ns	1320202
Appending int+float using FMT, pattern: %m%n/threads:6	509 ns	2883 ns	236166
Async, Sending int+float using MessageBuffer	1221 ns	1221 ns	573362
Async, Sending int+float using MessageBuffer/threads:6	579 ns	3458 ns	202482
Logging int+float using MessageBuffer, pattern: %d %m%n	1203 ns	1203 ns	584099
Logging int+float using MessageBuffer, pattern: %d %m%n/threads:6	982 ns	4584 ns	150492
Multiprocess logging int+float using MessageBuffer, pattern: %d %m%n	3496 ns	3496 ns	198909

[swebb2066]

The benchmarks on Windows 11 Visual Studio 2022 are:

Benchmark	Time	CPU	Iterations
Testing disabled logging request	3.41 ns	3.26 ns	263529412
Testing disabled logging request/threads:4	0.960 ns	3.84 ns	179200000
Appending 5 char string using MessageBuffer, pattern: %m%n	324 ns	315 ns	2036364
Appending 5 char string using MessageBuffer, pattern: %m%n/threads:4	293 ns	1116 ns	448000
Appending 49 char string using MessageBuffer, pattern: %m%n	396 ns	396 ns	1659259
Appending 49 char string using MessageBuffer, pattern: %m%n/threads:4	347 ns	1289 ns	400000
Appending int value using MessageBuffer, pattern: %m%n	697 ns	656 ns	1000000
Appending int value using MessageBuffer, pattern: %m%n/threads:4	541 ns	2128 ns	448000
Appending int+float using MessageBuffer, pattern: %m%n	1147 ns	1099 ns	640000
Appending int+float using MessageBuffer, pattern: %m%n/threads:4	812 ns	3139 ns	298668
Appending int value using MessageBuffer, pattern: [%d] %m%n	709 ns	698 ns	1120000
Appending int value using MessageBuffer, pattern: [%d] [%c] [%p] %m%n	767 ns	767 ns	1120000
Async, Sending int+float using MessageBuffer	1503 ns	1444 ns	497778
Async, Sending int+float using MessageBuffer/threads:4	1275 ns	4883 ns	224000
Logging int+float using MessageBuffer, pattern: %d %m%n	1394 ns	1381 ns	407273
Logging int+float using MessageBuffer, pattern: %d %m%n/threads:4	1951 ns	6810 ns	162908
Multiprocess logging int+float using MessageBuffer, pattern: %d %m%n	30028 ns	29681 ns	19478

[swebb2066]

This PR onlly reduced the overhead by about 4% on Ubuntu gcc.

Note that the benchmarks set BufferedIO=true

[swebb2066]

To improve performance on Windows, it would be necessary to use ReadDirectoryChangesW with a FILE_NOTIFY_CHANGE_FILE_NAME filter and poll for changes using WaitForMultipleObjects.

Seems a bit too much code to add given the level of usage of the appender.