Improving error handing when reading json config

[kheaactua]

Without mandatory fields set, the previous code would hit many exceptions in the happy load case. Thus, instead of depending on exceptions to detect the existence of keys in the parsed json, this change optionals to check to see whether a key exists. This should improve performance* (no stack unrolling for the exception) as well as allows some try/catch blocks to be removed (simpler code.)

Another driver for this change was that 4913ae3 swapped ellipses in the catch statements for std::exception. On the surface this seems like a positive change and even follows compiler warning message recommendations, however for unknown reasons on some platforms catch (std::exception) would catch the boost::property_tree::ptree_bad_path exceptions thrown by get_child, and on other platforms it wouldn't. Even though there is a proposal to revert that commit (eventually committed in 9259c09), reducing dependence on exceptions in favour of optionals obviates this issue.

Not every try/catch block was removed here, simply the ones where there's no conceivable change of an exception being thrown. The blocks surrounding the streamstream converters, and map access were left just in case.

Performance

(Measured in 3.5.5)

Using timing data from the log line Parsed vsomeip configuration in Xms, I saw the following improvements (sample size of n=1).

	qnx-baseline	qnx-change	aosp-baseline	aosp-change
Arith. Mean	41.73	10.64	5.84	3.14
Std.	47.32	5.97	4.98	3.14

Notes

• Formatted with clang-format
• I'm under the impression that upstream has converted to boost:json internally, so I'm not sure this PR will ever be accepted. In the meantime there it provides great speedups. For 3.5.6 for instance, there was a new block added in the config with the try/catch method, and while I don't know why, that caused one of our apps to slow down by 500ms on startup. I fixed up that block and the time it took to parse the config fell back to 9ms.