perf: restore preformatted blocks in one pass

[hownowstephen]

Each <pre> block was put back with strings.Replace(wrapped, prePlaceholder, block, 1). Every call scanned from the start and copied the whole text, so k blocks cost k full scans and k copies. restorePre now walks forward once with strings.Cut into a builder sized for the text plus the blocks.

Measured on documents of paragraphs, each followed by a <pre> block:

blocks     sec/op                  B/op                      allocs/op
2          9.41µ →  9.41µ    ~     14.7Ki →  14.1Ki   -4.7%   40 →  39
20        103.6µ →  86.1µ  -17%    268Ki  →  118Ki   -56%    215 → 196
200        2.16ms → 0.84ms -61%    15.0Mi →  1.0Mi   -93%    1,848 → 1,649

Most mail has no <pre> blocks, and the corpus documents are unchanged in both output and timing. The gain is on code-heavy mail such as release notes and developer newsletters. The benchmark was an ad hoc one and is not committed.

TestManyPreformattedBlocks checks that 50 blocks come back in order. Output is byte-for-byte identical to main across the corpus under every link style.