perf: refactor ignore logic to prevent duplicate work and allocations

[TimothyMakkison]

Builds on #1897, supports #1893

IgnoreList

• Convert alwaysIgnoredRules to use IgnoreRule[], avoiding the cost of compiling the regex for each construction of IgnoreList
• Instead of concatenating alwaysIgnoredRules and the parsed rules we simply evaluate one after the other to avoid allocating a List or larger array

IgnoreRule

• Use TrimStart(ReadOnlySpan<char> overload to avoid reallocating a string for every call
• Use preprocessor to use the more modern Split api that returns a Range instead of creating substrings.
  Related feat: refactored benchmarks and added CliBenchmarks #1896

StringExtensions

• Keeps the change from perf: remove string allocation using Replace(char overload #1897, avoids allocating a new string each call

Benchmarks - from #1896

Note that timing isn't very accurate due to low iteration count

Before

Method	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
FormatCli	737.4 ms	116.9 ms	30.36 ms	6000.0000	-	-	57.53 MB
CheckFiles	43,895.1 ms	639.7 ms	98.99 ms	53000.0000	52000.0000	2000.0000	459.28 MB

After

Method	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
FormatCli	643.6 ms	63.36 ms	16.46 ms	2000.0000	-	-	24.9 MB
CheckFiles	33,135.1 ms	842.61 ms	130.39 ms	45000.0000	44000.0000	2000.0000	393.78 MB

Pretty large performance gains for FormatCli, curious to see how large the real world impact will be.

[TimothyMakkison]

After #1888 it would be worth switching most of these methods over to using ReadOnlySpan<char> while using GetAlternateLookup.
In StringExtensions.NormalisePath should be changed to accept ReadOnlySpan<char>, checking for : either allocating a new string or returning a trimmed value.
There are other areas where SearchValues could replace string[] for lookups