Fix quadratic buffer growth in lossy WebP Vp8BitWriter

[gianlucaratta]

Prerequisites

• I have written a descriptive pull-request title
• I have verified that there are no overlapping pull-requests open
• I have verified that I am following the existing coding patterns and practice as demonstrated in the repository. These follow strict Stylecop rules 👮.
• I have provided test coverage for my change (where applicable)

Description

Fixes #3206

Vp8BitWriter.BitWriterResize compared the needed size against private readonly int maxPos, which is set to 0 in the constructor and never updated. The early return therefore never fired and ResizeBuffer(0, neededSize) computed the new size from a base of 0, i.e. neededSize rounded up to the next KiB. As a result the partition buffer was reallocated and copied every time the encoded output crossed a 1 KiB boundary (the very first flush even shrank the expected-size buffer down to 1 KiB), so allocations were quadratic in the encoded size.

libwebp updates bw->max_pos = new_size after resizing (src/utils/bit_writer_utils.c), and Vp8LBitWriter.BitWriterResize already grows from this.Buffer.Length. This PR removes the dead maxPos field and compares against / grows from this.Buffer.Length, so the buffer only reallocates when it is actually full and grows geometrically (1.5x, rounded up to the next KiB).

Flush is the only caller: it requests run + 1 bytes and then writes exactly run + 1 bytes starting at pos, so checking pos + extraSize <= Buffer.Length guarantees capacity for every write. NumBytes (pos) is unchanged, so WriteToStream / WriteToBuffer emit the same bytes.

Tests

• New Vp8BitWriterTests:
  • BitWriterResize_WithinCapacity_DoesNotReallocate: a resize request that fits the current buffer keeps the same array instance.
  • BitWriterResize_GrowsBufferGeometrically: writes 256 KiB through the writer and counts buffer replacements (12 with the fix, 256 before), and checks Buffer.Length >= NumBytes.
  • Both fail on main and pass with the fix.
• All WebP tests pass (dotnet test -c Release --filter "FullyQualifiedName~Webp": 444 passed, net10.0).
• Encoded output is byte-identical before/after: 42 files compared (the Lossy WebP encoder: Vp8BitWriter.maxPos is never updated, buffer is reallocated every 1 KiB (quadratic allocations) #3206 repro and a heavier noise image at 3 sizes, plus Calliphora.jpg, Bike.png, rgb-48bpp.png and alpha-blend-2.webp at Quality 10/75/100 x Method Fastest/Default/BestQuality).

Before / after (lossy, Quality = 90, default method, .NET 10, Release, Windows 11 x64; GC.GetTotalAllocatedBytes(true) around a single Save):

Repro from #3206:

Size	Output	Alloc. before	Alloc. after	Gen2 GCs	Time
1000x750	283 KiB	36.7 MiB	2.5 MiB	0 → 0	0.33 → 0.33 s
2000x1500	1,132 KiB	558.0 MiB	5.5 MiB	8 → 0	0.83 → 0.71 s
4000x3000	4,768 KiB	9,897.0 MiB	17.4 MiB	282 → 0	4.7 → 3.1 s

Noisier image (gradient + ±64 per-channel noise, larger output):

Size	Output	Alloc. before	Alloc. after	Gen2 GCs	Time
1000x750	465 KiB	99.9 MiB	3.0 MiB	2 → 0	0.41 → 0.35 s
2000x1500	1,848 KiB	1,548.5 MiB	7.4 MiB	34 → 0	1.10 → 0.77 s
4000x3000	7,401 KiB	24,797.9 MiB	24.9 MiB	686 → 0	7.6 → 3.0 s

[CLAassistant]

All committers have signed the CLA.