[.NET] Avoid allocation and improve parsing time

[obligaron]

🤔 What's changed?

This PR optimises allocations and speed by

• changing some types from class to struct
• removed/deferred some string trimming in GherkinLine
• change some yield return methods to explicit struct enumerators (no allocations & can be inlined)
• replaced arrays with IEnumerable so that List can be used, this avoids re-creating and coping with explicitly sized arrays

Results before:

Method	Runtime	FeatureFile	Mean	Error	StdDev	Gen0	Gen1	Allocated
Parser	.NET 8.0	tags.feature	23.01 μs	0.440 μs	0.556 μs	2.9297	0.1221	49.04 KB
ParserReuse	.NET 8.0	tags.feature	15.49 μs	0.215 μs	0.201 μs	2.5024	0.0916	41.18 KB
Parser	.NET Framework 4.8.1	tags.feature	31.99 μs	0.613 μs	0.656 μs	9.2773	0.4272	57.32 KB
ParserReuse	.NET Framework 4.8.1	tags.feature	23.45 μs	0.105 μs	0.088 μs	7.3853	0.2747	45.45 KB
Parser	.NET 8.0	very_long.feature	565.50 μs	5.492 μs	5.138 μs	95.7031	54.6875	1568.74 KB
ParserReuse	.NET 8.0	very_long.feature	546.43 μs	10.785 μs	10.088 μs	94.7266	47.8516	1558.77 KB
Parser	.NET Framework 4.8.1	very_long.feature	931.86 μs	5.545 μs	5.187 μs	261.7188	96.6797	1610.54 KB
ParserReuse	.NET Framework 4.8.1	very_long.feature	896.85 μs	2.761 μs	2.447 μs	259.7656	100.5859	1596.55 KB

Results after:

Method	Runtime	FeatureFile	Mean	Error	StdDev	Gen0	Gen1	Allocated
Parser	.NET 8.0	tags.feature	13.443 μs	0.0661 μs	0.0618 μs	1.5869	-	26.76 KB
ParserReuse	.NET 8.0	tags.feature	7.682 μs	0.0559 μs	0.0466 μs	1.1520	0.0305	18.86 KB
Parser	.NET Framework 4.8.1	tags.feature	20.678 μs	0.0940 μs	0.0879 μs	5.2185	0.2136	32.13 KB
ParserReuse	.NET Framework 4.8.1	tags.feature	13.165 μs	0.0836 μs	0.0782 μs	3.2806	0.1068	20.21 KB
Parser	.NET 8.0	very_long.feature	295.813 μs	3.1228 μs	2.6077 μs	49.8047	30.2734	816.38 KB
ParserReuse	.NET 8.0	very_long.feature	288.504 μs	0.9828 μs	0.8713 μs	49.3164	25.3906	806.37 KB
Parser	.NET Framework 4.8.1	very_long.feature	564.142 μs	3.3668 μs	3.1493 μs	138.6719	54.6875	853.5 KB
ParserReuse	.NET Framework 4.8.1	very_long.feature	540.853 μs	3.3997 μs	3.1801 μs	135.7422	45.8984	839.47 KB

⚡️ What's your motivation?

This reduces parsing time by ~45% and allocations by ~45%.

🏷️ What kind of change is this?

• 🏦 Refactoring/debt/DX (improvement to code design, tooling, etc. without changing behaviour)
• 💥 Breaking change (incompatible changes to the API)

♻️ Anything particular you want feedback on?

The removed IGherkinLine interface is used in the Reqnroll VisualStudio plugin (I haven't found other implementations on github). But the custom implementation HotfixLine is probably not needed anymore. At least the added regression tests will pass with the original IGherkinLine implementation from this repository.

📋 Checklist:

• I agree to respect and uphold the Cucumber Community Code of Conduct
• Users should know about my change
  • I have added an entry to the "Unreleased" section of the CHANGELOG, linking to this pull request.