This is preparation for the work required to enable IUtf8SpanParsable support on IPAddress and IPNetwork.
Also replaced a few instances of unsafe code with spans.
This uses the same style of generic constraints as the numeric parsing in corelib (although it can't access IUtfChar<TSelf>.)
All tests pass, although the UTF8 code paths haven't yet been tested.
Some restructuring is pending as I plumb IPAddressParser to the updated generic methods.
Future commits will handle the downstream dependencies of IPvXAddressHelper - at present, this looks like Uri and TargetHostNameHelper.

This allows me to eliminate the nested class and now-redundant generic type constraints on each method

These are just the changes required to make System.Private.Uri compile.
* Split IPAddressParser and IPv6AddressHelper into multiple files and shared common components between projects.
* Updated some of the Uri-specific additions to IPvXAddressHelper files to use spans.
* Minor adjustments to project files to support the above.

Removed a using statement in the ref project, small reduction in the csproj diff for System.Net.Primitives

Moved the generic type definition from the IP*AddressParser classes to individual methods, and moved the constant-value fields to local variables.
Swapped out the arrays in these constant-value fields to a ReadOnlySpan to reduce allocations.
Replaced calls to TChar.CreateTruncating with TChar.CreateChecked.

Also made one adjustment in Uri so that it uses BinaryPrimitives to convert a long to a Span<byte>, rather than manual bit shifts.

Renamed bytesConsumed parameter to charsConsumed.
`IPNetwork.TryParse(ReadOnlySpan<byte>, IPNetwork)` now finds the last index of a '/' rather than the first. This syncs its behaviour with the `ReadOnlySpan<char>` equivalent.
Removed unnecessary type constraints which lingered after the removal of the class-level generic type constraint.

This is only used when handling an IPv4 address embedded inside an IPv6 address, but this can only appear once, so the modification is unnecessary

* Completed rename process of bytesConsumed to charsConsumed.
* Replaced remaining reference of TChar.CreateChecked with CreateTruncating.
* Removed unnecessary nullability annotation.
* Formatting change

* Removed use of helper method for number parsing.
* Removed redundant check of parts[0] in IPv4AddressHelper.
* Added basic heuristic to IPv6AddressHelper to avoid searching for an embedded IPv4 address if the string doesn't contain a '.'.

IPAddressParser.Common.cs is no longer needed. This should also clean up the PR diff

* Slightly reducing memory usage by storing the parts as uints.
* Removing unnecessary assignment of the maximum digit value for octal, deferring another.
* Removed case-insensitivity in the hexadecimal parsing.
* Remove unnecessary bitmasks in final assembly of address - values have already been verified as < 0xFF.

* Reverted from Spans to pointers

* Hoisted a condition out of its loop.
* Switched from a series of comparisons when validating a hex digit to a simple table lookup.
* Adjusted IPv6AddressHelper.Parse to eliminate range checks on the components. Nothing reaches .Parse without being validated, so there's no overflow risk here.

Mostly in the IPv6 parsing.
* Removed HexConverter, which adds a little more latency than expected.
* For short (<= 16) IPv6 addresses, ROS.Contains and ROS.IndexOf are more expensive than iterating over each character.

Style/comment changes to IPv4AddressHelper.
Swapped IPv6AddressHelper back to the previous switch block.

* Removed references to unchecked.
* Changed numeric character detection to align with char.IsAsciiDigit.
* Changed hex character detection to use HexConverter.
* Cosmetic change: reordered if-condition and changed comment to match - slightly cleans up diff.
* Cosmetic change: line spacing.

There are now no circumstances where a string is passed to if_nametoindex or SystemNative_InterfaceNameToIndex, or the calling InterfaceNameToIndex method in InterfaceInfoPal. Removed these.

An interface name comes from a scope in a valid IPv6 address. There'll always be at least three characters (::%) in front of the scope, so taking the length of an interface name which stripping these three leading characters away and adds a single null terminator will never overflow an integer.

This eliminates a number of changes required to transcode from UTF8 and Unicode to ANSI.
Instead of using if_nametoindex on Windows, this now uses ConvertInterfaceNameToLuidW & ConvertInterfaceLuidToIndex.

When running in a container, Windows does not have a static loopback interface name.
Moved tests of InterfaceInfoPal to the PalTests project, rather than testing with hardcoded IP addresses with real scope IDs.
Made the tests use real (and in Windows' case, dynamic) interface names.

* SetLastError has the correct value in the Windows interop layer.
* Preserve InterfaceNameToIndex's errno in the Unix InterfaceInfoPal.

* Replaced int.CreateTruncating with a JIT pattern which converts from TChar to a ushort.
* Removed various TChar.CreateTruncating constants, reinstating direct character constants.
* Replaced branch statements when parsing IPv4 numerics with a HexConverter lookup.
* IPv4 parts stored in an array of longs to improve register usage.
* Small improvement in reassembly to eliminate extra array access.
* Reverted changes to Uri parsing layer, removing the translation logic between a Span and a pointer for IPv4 addresses.

…ss-ipnetwork

* Replaced ushorts with integers to remove some zero-extensions.
* Replaced NativeMemory with an ArrayPool<byte>.
* Removed unnecessary assignment to ch.