Fix four fidelity losses when converting an existing Mudlet package

[TheLastDarkthorne]

Four independent fidelity fixes, found while converting an existing Mudlet package into a
muddler project. Each is small, self-contained and revertable on its own.

Fixes #52
Fixes #53
Fixes #54
Fixes #55

The commits

Keep declared folder attributes when merging with directory placeholders (#52)

createItems() synthesises a placeholder item for each directory in a path, carrying only
the name and default attributes. mergeDown() then merges the placeholder with the real
declaration by name, but keeps whichever it visited first and only concatenates children,
so the declaration's attributes were dropped when the placeholder came first. Synthesised
placeholders are now marked, and a real declaration wins the merge; child order is
unchanged.

Honour multilineDelta for triggers that are not multiline (#53)

conditonLineDelta is Mudlet's line window for a chained trigger, which applies whether or
not the trigger is multiline — its UI exposes the spinner independently of the multiline
checkbox. It was only copied when multiline was "yes".

Report a usable error for a malformed colour pattern (#55)

Splitting on , and indexing [1] unconditionally turned any other form into an
ArrayIndexOutOfBoundsException naming neither the trigger nor the pattern. It now says
which trigger and what it received.

Accept explicit keyCode and keyModifier on keys (#54)

A keys string like "ctrl+F1" is convenient when authoring by hand, but a package
imported from Mudlet XML carries raw Qt codes and had no way to express them, so imported
keys lost their binding. keyCode/keyModifier are used when given, falling back to
parsing keys as before.

Not included

#56 (same-named siblings being dropped) has no fix here. mergeDown/fullMerge are built
around matching by name, and separating "directory placeholder that should merge" from "two
real items that should not" looks like a design decision rather than a small fix — happy to
attempt it given a steer on the preferred approach.

Verification

Runnable reproductions for all five issues, each with a passing control:
https://github.com/TheLastDarkthorne/mudlet-tooling-repros

Every case was run against the stock 1.1.0 release jar and against this branch; the four
fixed cases change behaviour and the control cases do not.

Beyond the minimal cases, this branch was exercised on the conversion that turned these up:
svof — 25 Mudlet modules, ~2,500
triggers, 120 aliases, 70 scripts. Each module was converted to a muddler project, built,
and the generated XML compared against the original
object-by-object (structure, names, scripts, patterns and pattern types, event handlers,
and the behavioural flags). All 25 modules round-trip with no differences using this branch.

On stock 1.1.0 the same comparison loses the isActive state of 8 trigger groups, zeroes 30
multilineDelta values, drops the package's only key binding, and cannot build at all
without working around the colour-pattern crash.

Note on the build

This branch is built and tested with the wrapper's own Gradle 7.5, so it carries no build
changes. (Gradle 7.5 cannot parse class file major version 63, so building on JDK 19+ needs
a newer wrapper — kept out of this PR as an unrelated concern, and verified separately to
produce byte-identical output.)

[TheLastDarkthorne]

Forgot to mention, I'm converting the in-client-svof branch of svof in case you want to test on it later. I used demuddler on it (which also required some fixes on its own to work on there). I will be pushing a repo with the end result (with the fixes) soon.