Reduce HTTP headers validation overhead

[MihaZupan]

When adding/reading headers where we don't have a special parser, "parsing" only validates that there are no new lines in the value. This change special-cases this (common) case and avoids allocating the HeaderStoreItemInfo.

Existing code paths where only non-validating APIs are used stay the same.
Overhead for cases of reading headers with validation, where they were added without validation, and we do have a known parser (still common) is minimal and an acceptable tradeoff IMO (extra branch).

Method	Toolchain	Mean	Error	Ratio	Allocated	Alloc Ratio
Add	main	44.31 ns	0.386 ns	1.00	32 B	1.00
Add	pr	21.35 ns	0.018 ns	0.48	-	0.00
AddEnumerable	main	36.27 ns	0.592 ns	1.00	32 B	1.00
AddEnumerable	pr	28.80 ns	0.265 ns	0.79	-	0.00
GetValues	main	92.39 ns	0.302 ns	1.00	64 B	1.00
GetValues	pr	40.35 ns	0.090 ns	0.44	32 B	0.50
AddAndGetValues	main	94.01 ns	0.192 ns	1.00	64 B	1.00
AddAndGetValues	pr	42.92 ns	0.229 ns	0.46	32 B	0.50
CloneHeaders	main	905.95 ns	1.891 ns	1.00	1112 B	1.00
CloneHeaders	pr	490.74 ns	1.064 ns	0.54	600 B	0.54

Benchmark code

BenchmarkRunner.Run<HeadersBench>(args: args);

[MemoryDiagnoser(false)]
public class HeadersBench
{
    private readonly HttpResponseHeaders _headers = new HttpResponseMessage().Headers;
    private readonly HttpRequestHeaders _headersToClone = new HttpRequestMessage().Headers;
    private readonly string[] _fooAsArray = ["Foo"];

    public HeadersBench()
    {
        _headersToClone.TryAddWithoutValidation("priority", "u=0, i");
        _headersToClone.TryAddWithoutValidation("sec-ch-ua-mobile", "?0");
        _headersToClone.TryAddWithoutValidation("sec-ch-ua-platform", "\"Windows\"");
        _headersToClone.TryAddWithoutValidation("sec-fetch-dest", "document");
        _headersToClone.TryAddWithoutValidation("sec-fetch-mode", "navigate");
        _headersToClone.TryAddWithoutValidation("sec-fetch-site", "none");
        _headersToClone.TryAddWithoutValidation("sec-fetch-user", "?1");
        _headersToClone.TryAddWithoutValidation("upgrade-insecure-requests", "1");
    }

    [Benchmark]
    public void Add()
    {
        _headers.Add("X-Custom", "Foo");
        _headers.Clear();
    }

    [Benchmark]
    public void AddEnumerable()
    {
        _headers.Add("X-Custom", _fooAsArray);
        _headers.Clear();
    }

    [Benchmark]
    public object GetValues()
    {
        _headers.TryAddWithoutValidation("X-Custom", "Foo");
        IEnumerable<string> values = _headers.GetValues("X-Custom");
        _headers.Clear();
        return values;
    }

    [Benchmark]
    public object AddAndGetValues()
    {
        _headers.Add("X-Custom", "Foo");
        IEnumerable<string> values = _headers.GetValues("X-Custom");
        _headers.Clear();
        return values;
    }

    [Benchmark]
    public HttpRequestHeaders CloneHeaders()
    {
        HttpRequestHeaders newHeaders = new HttpRequestMessage().Headers;

        foreach (KeyValuePair<string, IEnumerable<string>> header in _headersToClone)
        {
            newHeaders.Add(header.Key, header.Value);
        }

        return newHeaders;
    }
}

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull Request Overview

This PR optimizes header handling by bypassing the full parser for headers without a known parser, special‐casing single‐value collections to avoid allocations, and updating both TryGetValues and the enumerator to reduce heap traffic.

• Bypass parser and allocation when descriptor.Parser is null in both single- and multi-value Add overloads
• Short-circuit single-element IEnumerable<string?> to reuse the single-value Add path
• Optimize TryGetValues and GetEnumeratorCore to return raw string values without allocating HeaderStoreItemInfo

[CarnaViire]

LGTM, thanks!