Remove some branches in OptimizedInboxTextEncoder.GetIndexOfFirstCharToEncode

[MihaZupan]

Follow-up to review feedback on #129625 (comment).

Since idx is always in range (either 0 when the SIMD search is skipped, or
a value already validated against data.Length), the (uint)idx < (uint)data.Length
guard and the separate remaining span are unnecessary. We now slice data in
place inside the proven-in-range branch and run the loops directly over it.

No behavior change. Codegen for the method improves:

	Before	After
Code size	516 bytes	485 bytes
Instructions	156	146
Callee-saved regs spilled	+1 (r14)	—

The redundant range check is gone and the extra length register is no longer
needed. Hot-loop bounds checks remain eliminated, and the data.Slice(idx)
compiles to plain pointer arithmetic with no ThrowArgumentOutOfRange.

Validated: System.Text.Encodings.Web.Tests (610 tests) pass.

Note

This PR was authored with assistance from GitHub Copilot.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

This PR simplifies OptimizedInboxTextEncoder.GetIndexOfFirstCharToEncode(ReadOnlySpan<char>) by removing a redundant range-check / secondary “remaining” span and instead slicing data in-place once idx is proven in-range, aiming to improve generated code quality without changing behavior.

Changes:

• Replace the (uint)idx < (uint)data.Length guard + remaining span with in-place data = data.Slice(idx) and direct iteration over data.
• Keep the 8x-unrolled scan and fall back to a foreach for the tail.
• Adjust the return flow to return -1 on full consumption and return idx on first disallowed char.