Allow ROW::CopyRangeFrom to be vectorized (#15267)

By rewriting the first major copy loop in `CopyRangeFrom` to use
pointers/iterators instead of indices for iteration, the autovectorizer
kicks in end neatly rewrites it as an unrolled SIMD loop. This improves
performance during traditional window resizes by roughly 2x and will
be quite helpful in the future for our more complex reflow resize.

Unfortunately, MSVC unrolls the loop by 4x which is too much for our
purpose, but there's no option to change that. It's still better than
not having any vectorization however, since it kicks in at 32 columns.

It also renames the function to `CopyTextFrom` be more in line with
the others and to avoid confusion, because it doesn't copy attributes.

## Validation Steps Performed
* Traditional resizing works ✅

.github/actions/spelling/expect/expect.txt

@@ -2062,6 +2062,7 @@ vcpkg
 vcprintf
 vcxitems
 vec
+vectorize
 vectorized
 VERCTRL
 VERTBAR

src/buffer/out/Row.cpp

@@ -631,26 +631,36 @@ catch (...)
     const auto baseOffset = til::at(charOffsets, 0);
     const auto endOffset = til::at(charOffsets, colEndInput);
     const auto inToOutOffset = gsl::narrow_cast<uint16_t>(chBeg - baseOffset);
+#pragma warning(suppress : 26481) // Don't use pointer arithmetic. Use span instead (bounds.1).
+    const auto dst = row._charOffsets.data() + colEnd;
 
-    // Now with the `colEndInput` figured out, we can easily copy the `charOffsets` into the `_charOffsets`.
-    // It's possible to use SIMD for this loop for extra perf gains. Something like this for SSE2 (~8x faster):
-    //   const auto in = _mm_loadu_si128(...);
-    //   const auto off = _mm_and_epi32(in, _mm_set1_epi16(CharOffsetsMask));
-    //   const auto trailer  = _mm_and_epi32(in, _mm_set1_epi16(CharOffsetsTrailer));
-    //   const auto out = _mm_or_epi32(_mm_add_epi16(off, _mm_set1_epi16(inToOutOffset)), trailer);
-    //   _mm_store_si128(..., out);
-    for (uint16_t i = 0; i < colEndInput; ++i, ++colEnd)
-    {
-        const auto ch = til::at(charOffsets, i);
-        const auto off = ch & CharOffsetsMask;
-        const auto trailer = ch & CharOffsetsTrailer;
-        til::at(row._charOffsets, colEnd) = gsl::narrow_cast<uint16_t>((off + inToOutOffset) | trailer);
-    }
+    _copyOffsets(dst, charOffsets.data(), colEndInput, inToOutOffset);
 
+    colEnd += colEndInput;
     colEndDirty = gsl::narrow_cast<uint16_t>(colBeg + colEndDirtyInput);
     charsConsumed = endOffset - baseOffset;
 }
 
+#pragma warning(push)
+#pragma warning(disable : 26481) // Don't use pointer arithmetic. Use span instead (bounds.1).
+[[msvc::forceinline]] void ROW::WriteHelper::_copyOffsets(uint16_t* __restrict dst, const uint16_t* __restrict src, uint16_t size, uint16_t offset) noexcept
+{
+    __assume(src != nullptr);
+    __assume(dst != nullptr);
+
+    // All tested compilers (including MSVC) will neatly unroll and vectorize
+    // this loop, which is why it's written in this particular way.
+    for (const auto end = src + size; src != end; ++src, ++dst)
+    {
+        const uint16_t ch = *src;
+        const uint16_t off = ch & CharOffsetsMask;
+        const uint16_t trailer = ch & CharOffsetsTrailer;
+        const uint16_t newOff = off + offset;
+        *dst = newOff | trailer;
+    }
+}
+#pragma warning(pop)
+
 [[msvc::forceinline]] void ROW::WriteHelper::Finish()
 {
     colEndDirty = row._adjustForward(colEndDirty);

src/buffer/out/Row.hpp

@@ -171,6 +171,7 @@ class ROW final
         void ReplaceCharacters(til::CoordType width) noexcept;
         void ReplaceText() noexcept;
         void CopyTextFrom(const std::span<const uint16_t>& charOffsets) noexcept;
+        static void _copyOffsets(uint16_t* dst, const uint16_t* src, uint16_t size, uint16_t offset) noexcept;
         void Finish();
 
         // Parent pointer.