This change reverts 0ad4342 and increases to 255 the length of strings
that can be passed directly to Rope≫Append() without getting truncated.
Otherwise, without this change, any string passed directly to
Rope≫Append() whose length exceeds a particular system-dependent limit
(which in practice seems to be 15) will get silently truncated.

Relates to #153.

This change makes the DOM implementation append strings directly when
setting attribute values and when appending text nodes and comments.
That in turn allows passing non-constant strings to Element≫AppendChild()
and Document≫AppendChild() and Element≫SetAttribute().

Otherwise, without this change, the strings we use for attribute values and
text nodes and comments either must be constants, or else we have to create
CutRopes, append strings to them, and then use TText.CreateDestructively()
(to create text nodes and comments) and Element≫SetAttributeDestructively()
(to set attribute values).

… directly

…vely

Currently the longest string we’re appending directly to Ropes is the
40-character SourceGitSHA string for the snapshots, and the second-
longest string is the 30-character “This section is non-normative.”.

This change switches that SourceGitSHA-handling code to instead using
Scratch.Append(@SourceGitSHA) and TText.CreateDestructively(Scratch) —
so that we can reduce down to 30 the size of ShortStrings we allow to be
appended directly to Ropes.

The larger the size of ShortStrings we allow to be appended directly to
Ropes, the more memory Wattsi consumes at runtime.

We don’t need the size to be anywhere near as big as 255 — the existing
code never actually needs to directly append strings with lengths any
longer than 30. If we ever do run into need to append string lengths
longer than 30, we could at that time just bump UTF8InlineSize up to
whatever new size we actually need.

Upping the size to 30 from the old size of 15 seems to increase the
memory consumption by about 15–20%, or around 80–100MB. (In comparison,
upping it to 255 seems to roughly double the memory consumption.)