perf(json5): Remove use of serde(untagged)

[epage]

Prep for #472

serde(untagged) translates to an if/else if ladder that ends up being pretty expensive, including the error generation for each branch that gets skipped. This switches us to dispatch on type instead, speeding things up.

This is also prep for speeding up builds by removing the need for serde_derive. Once serde_core is published, we can depend on that and not wait for syn, quote, etc.

[coveralls]

Pull Request Test Coverage Report for Build 16917008830

Details

• 10 of 10 (100.0%) changed or added relevant lines in 1 file are covered.
• 2 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.1%) to 65.424%

---

Files with Coverage Reduction	New Missed Lines	%
src/error.rs	1	71.08%
src/format.rs	1	40.0%

Totals
Change from base Build 16806457660:	0.1%
Covered Lines:	965
Relevant Lines:	1475

---

💛 - Coveralls

[dtolnay]

Since this PR is partially motivated by performance, it is worth knowing about dtolnay/erased-serde#61 (which is used by serde-untagged). For very deeply nested config you would see O(n²) performance in the new implementation compared to O(n) in the previous one. Ultimately the best performance would be from a more verbose impl that resembles https://github.com/serde-rs/json/blob/v1.0.142/src/value/de.rs#L22, which completely eliminates the quadratic factor and also reduces linear factor from heap allocations inside erased-serde and constant factor from heap allocations inside UntaggedEnumVisitor.