Add undo/redo and fix lost edits in the scripting terminal

[rubensworks]

Phase 2 of the scripting terminal UX improvements, stacked on #77, so review and merge that one first. Each commit fixes one problem.

I reproduced every problem in a dev client driven by clientdevbridge (or with a unit test that fails on the old code) before fixing it, then checked it again with this branch.

Commit	Problem before	Result with this PR
Add undo and redo to the script editor	A right-click selected the whole 40-line script; one key press replaced it, and that was saved to disk (z). No undo existed.	Right-click no longer selects everything. Ctrl+Z undoes and Ctrl+Y / Ctrl+Shift+Z redo, with consecutive typing grouped into one step (checked: abc, Enter, de undo as 3 steps). History is kept per file. Unit tests added.
Don't echo synced scripts back to the client	The server sent each synced script back. A keystroke typed between the sync and the echo was overwritten: the editor showed aAB but aA was saved.	aAB is saved.
Write scripts to disk atomically	The server's file watcher could reload a half-written file into memory while the server was writing it.	New unit test that saves a 1 MB script 20 times while the watcher runs. It fails on the old code with partial contents.
Show external script changes in the open editor	After editing the file outside the game, the open editor kept the old text, and the next keystroke wiped the external change from disk.	The editor reloads the new text as an undoable step and keeps the cursor. The external change survives further typing.
Keep file permissions when writing scripts atomically	The atomic write used Files.createTempFile, which creates files readable only by the owner, so flushed scripts became rw-------.	New unit test that fails without the fix (rw-------) and passes with it: new files get rw-r--r--, and existing files keep their permissions.
Only apply external changes when the file watcher reloads scripts	The watcher also fires on the server's own writes, and its reload replaced everything in memory with what was on disk at that moment. Right after creating a new disk's folder, before the file was written, this made the first script disappear. Two concurrent reloads could also overwrite each other.	The server remembers what it last wrote or read per file, and applies only files that changed outside the mod. Reloads, flushes and script changes now take a shared lock.
Notice all externally created script files	Files created outside the game didn't show up until some other file changed. Each reload cancelled and re-registered the folder watchers, losing events in between, and only folders that already contained scripts were watched. Reproduced in-game: deleting 4 scripts and creating 3 new files (one in a new lib/helpers/ folder) in one shell command left the terminal showing the old list.	All folders of a disk are watched, including empty and intermediate ones, before its files are read; watchers are no longer re-registered; deleted folders are cleaned up. New tests (deletes followed by creates; files in new nested, intermediate and empty folders) fail 3/3 on the old code and pass now. Rechecked in-game: the terminal list matches the disk right after the same batch, and after adding files to an empty and an intermediate folder.

About the watcher reload commit

I found it because the partial-contents test was flaky: run 15 times in one JVM, it failed 13 times, always observing the script as null. That was this race, not partial contents. After both watcher commits, 30 runs of the whole ScriptingDataTest class (17 tests) had no failures. There are tests for the reload during the first flush (fails on the old code), and for external modifications and deletions still being applied.

This race is also a plausible cause of the "empty script after an external write" case I saw once earlier, but I didn't reproduce that again to confirm.

Why the atomic write commit is needed

Without it, the external-changes commit caused a regression: typing hello world into a new file gave orld, o worldllhe and rldhello wo, because the editor applied the half-written content that the watcher had read back. With the atomic write, all four test files came out exactly hello world.

Testing

• ./gradlew build passes, including the new unit tests.
• ./gradlew runGameTestServer: all 22 tests pass.

🤖 Generated with Claude Code

https://claude.ai/code/session_018MRyTdV3fSV5Q5NPGep4q9

[github-actions]

Coverage Report for CI Build 39131475676

Coverage decreased (-0.3%) to 46.454%

Details

• Coverage decreased (-0.3%) from the base build.
• Patch coverage: No coverable lines changed in this PR.
• 481 coverage regressions across 8 files.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

481 previously-covered lines in 8 files lost coverage.

File	Lines Losing Coverage	Coverage
org/cyclops/integratedscripting/client/gui/component/input/WidgetTextArea.java	215	0.0%
org/cyclops/integratedscripting/client/gui/container/ContainerScreenTerminalScripting.java	132	0.0%
org/cyclops/integratedscripting/inventory/container/ContainerTerminalScripting.java	64	26.67%
org/cyclops/integratedscripting/core/network/ScriptingData.java	41	78.95%
org/cyclops/integratedscripting/client/gui/component/input/TextFieldHelperExtended.java	17	0.0%
org/cyclops/integratedscripting/core/network/GraalScriptFactory.java	5	60.53%
org/cyclops/integratedscripting/network/packet/TerminalScriptingModifiedScriptPacket.java	4	0.0%
org/cyclops/integratedscripting/core/evaluate/ScriptVariable.java	3	81.25%

---

Coverage Stats

Relevant Lines:	2761
Covered Lines:	1405
Line Coverage:	50.89%
Relevant Branches:	976
Covered Branches:	331
Branch Coverage:	33.91%
Branches in Coverage %:	Yes
Coverage Strength:	23.27 hits per line

---

💛 - Coveralls

[rubensworks]

Replaced by #85: same changes, rebased onto master-1.21-lts with every commit as one standalone change.

---

Generated by Claude Code