Use the span based Convert APIs in the Base64 polyfill

[SimonCropp]

The Base64 polyfill (net11 members) encoded and decoded through source.ToArray() and intermediate strings/arrays on every target.

On NETCOREAPP2_1_OR_GREATER || NETSTANDARD2_1_OR_GREATER the helpers now use the span based BCL APIs:

• EncodeToString -> Convert.ToBase64String(ReadOnlySpan<byte>)
• EncodeToChars(source, destination) / TryEncodeToChars -> Convert.TryToBase64Chars straight into the destination
• EncodeToChars(source) -> Convert.TryToBase64Chars straight into the result array
• DecodeFromChars / TryDecodeFromChars / DecodeFromUtf8(source) -> Convert.TryFromBase64Chars into a stack (up to 256 bytes) or pooled buffer

On every target EncodeToUtf8(source) now calls Base64.EncodeToUtf8 straight into a right-sized array instead of building a string and narrowing it.

The contract is unchanged:

• invalid input still throws FormatException. Convert.TryFromBase64Chars returns false for both invalid input and a destination that is too small, so the decode goes to a buffer of source.Length / 4 * 3 bytes (the most that valid input can produce); a false then can only mean invalid input, and the array overload is called to throw the same exception as before
• a destination that is too small still returns false with 0 written (or throws ArgumentException from DecodeFromChars/EncodeToChars), invalid input still wins over a destination that is too small, and the destination is still only written to on success
• whitespace is still skipped

Targets without the span based Convert APIs (net4x, netstandard2.0, netcoreapp2.0) keep the previous code for everything except EncodeToUtf8(source). The OperationStatus overloads are not touched here.

Tests

• every changed overload against Convert for every length from 0 to 400 (all remainders, stack and pooled buffers), with a sentinel after the expected output
• destinations one element too small for the Try and throwing overloads
• empty and whitespace-only input, whitespace inside the input
• invalid input (bad character, bad length, bad padding; short and long enough to take the pooled path) for all decode overloads, including together with a destination that is too small
• non-ASCII bytes in DecodeFromUtf8

The real net11 Base64 reports partial progress when the destination is too small and reports that ahead of later invalid input; the tests check those two points only where the polyfill is in use (runtime check), everything else runs against both.

Measurements

Stopwatch / GC.GetAllocatedBytesForCurrentThread, Release, minimum of 4 alternating runs of before and after (each the best of 9 rounds after warm-up).

net10.0

	before	after
EncodeToString(32B)	30 ns / 168 B	25 ns / 112 B
EncodeToString(1KB)	222 ns / 3,808 B	189 ns / 2,760 B
EncodeToChars(32B, dest)	41 ns / 168 B	15 ns / 0 B
EncodeToChars(1KB, dest)	326 ns / 3,808 B	83 ns / 0 B
EncodeToChars(1KB)	443 ns / 6,568 B	230 ns / 2,760 B
TryEncodeToChars(1KB, dest)	327 ns / 3,808 B	148 ns / 0 B
EncodeToUtf8(32B)	45 ns / 240 B	16 ns / 72 B
EncodeToUtf8(1KB)	661 ns / 5,200 B	110 ns / 1,392 B
DecodeFromChars(32B)	46 ns / 168 B	40 ns / 56 B
DecodeFromChars(1KB)	1,018 ns / 3,808 B	676 ns / 1,048 B
DecodeFromChars(1KB, dest)	1,158 ns / 3,808 B	538 ns / 0 B
TryDecodeFromChars(1KB, dest)	1,060 ns / 3,808 B	537 ns / 0 B
DecodeFromUtf8(1KB)	1,273 ns / 3,808 B	1,176 ns / 1,048 B

net48 (only EncodeToUtf8(source) changes)

	before	after
EncodeToUtf8(32B)	205 ns / 248 B	63 ns / 72 B
EncodeToUtf8(1KB)	1,783 ns / 5,208 B	715 ns / 1,392 B