fix(run): don't print the built-in variables setup for .cfm templates

[bpamiri]

Upstream issue: cybersonic#137

What

lucli run hello.cfm (with <cfoutput>hello #1+1#</cfoutput>) printed about 20 lines of setup code (// LuCLI Built-in Variables Setup, __scriptDir = '…', …, // === Original Script Content ===) before hello 2. .cfs files were not affected.

Why

executeTemplate (the .cfm path) prepended the built-in variables setup, which is CFML script, to the template as plain text. In a template, text outside <cfscript> is output.

How

• executeTemplate builds the content with a new wrapBuiltinSetupForTemplate(setup, template): <cfscript> + setup + </cfscript>, then the template unchanged, with nothing in between. So the setup runs and the output is exactly what the template renders.
• The marker comment is not added for templates.
• injectBuiltinVariables is unchanged for any other caller.
• .cfs scripts go through executeScriptContent and are untouched.

Tests

• New: TemplatePreludeTest. The setup is inside <cfscript>, the template follows with nothing in between, there's no marker, and an empty or null setup leaves the template unchanged.

• End to end (isolated LUCLI_HOME / -Duser.home):

  	v0.6.2.2	this branch
  run hello.cfm	the setup lines, then hello 2	hello 2
  run vars.cfm (<cfoutput>#listLast(__scriptPath, "/")#</cfoutput>)	vars.cfm	vars.cfm

  So the variables are still available.

• Full suite (mvn test -DargLine="-Duser.home=<empty dir>"): 771 tests. The only failures are the 17 known clean-home failures that base wheels shows under the same sandbox (Test suite isn't isolated from the developer's LuCLI home: writes into ~/.lucli, fails in a clean home #25); none are new. The real ~/.lucli was untouched. Unsandboxed, CI runs the normal way.

Fork bookkeeping

CHANGELOG ## Unreleased entry; FORK.md "Next queue" row 13. It will conflict on those two files with whichever of #22–#24 merges first; I'll merge wheels in when it's this PR's turn.

Upstream

The same change goes to cybersonic/LuCLI as a cherry-pick -x onto upstream main (without FORK.md) once this is reviewed.

🤖 Generated with Claude Code

[bpamiri]

rev1-r3 review @ c26936d: code APPROVE (final after the wheels merge + CI on that head)

End to end (the RED/GREEN): JDK 21 jars, isolated LUCLI_HOME + -Dlucli.home, sandboxed HOME + -Duser.home, lucli run <file>.

File	base (wheels @ 8942048)	this branch
hello.cfm: <cfoutput>hello #1+1#</cfoutput>	19 lines: // LuCLI Built-in Variables Setup, __scriptDir = '…', … then hello 2	hello 2 (1 line)
vars.cfm: #listLast(__scriptPath,"/")#|#arrayLen(__arguments)#	setup text, then the values	vars.cfm|0: the built-ins are still set
s.cfs: writeOutput("cfs " & (1+2))	cfs 3	cfs 3 (the .cfs path is untouched)

Nothing was written under the sandboxed user.home in either run. TemplatePreludeTest passes 2/0/0. It can't be RED on base, because wrapBuiltinSetupForTemplate is new; the table above is the behavioural RED.

Code: minimal. The setup runs inside <cfscript>…</cfscript> with the template appended directly, so nothing between them is output. There's no marker comment, an empty or null setup returns the template unchanged, and injectBuiltinVariables stays as it is for every other caller.

Nits, not blocking:

• The .cfm path no longer wraps the setup in Timer.start/stop("injectBuiltinVariables"), so --timing loses that span for templates. Consider keeping it for parity.
• The verbose and debug dump right after still prints === CFS Script with Built-in Variables === for a .cfm. That's cosmetic.

Merge state: BLOCKED/conflicting on CHANGELOG/FORK.md as expected. I'll delta-check the wheels merge (only those two files) and read CI on that head before the final.

[github-actions]

Integration Tests (ubuntu-latest, Java 21)

78 tests   78 ✅  1m 23s ⏱️
14 suites   0 💤
 1 files     0 ❌

Results for commit 2ce882e.

[github-actions]

Unit Tests (ubuntu-latest, Java 21)

 64 files   64 suites   2m 4s ⏱️
780 tests 774 ✅ 6 💤 0 ❌
785 runs  779 ✅ 6 💤 0 ❌

Results for commit 2ce882e.

[bpamiri]

rev1-r3 review @ 2ce882e: APPROVE (final)

There are two commits on top of the approved c26936d:

• ed0b7a3, a merge of wheels (4773b61, which includes fix(modules): pass --env / --envfile through the module shortcut #21–fix(modules): root flags before a reserved subcommand under an alias #24). The PR's src/ changes against wheels at the merge are identical to c26936d, and the CHANGELOG/FORK.md resolution keeps rows 9–13 in order.
• 2ce882e addresses both of my nits. The .cfm path wraps the setup in Timer.start/stop("injectBuiltinVariables"), and the stop is in a finally, so the span closes even when the setup throws. That's slightly more robust than injectBuiltinVariables itself. The verbose/debug header now reads === CFM Template with Built-in Variables ===. No other change.

PR CI on this head: 8 pass (Build ×3, Unit ×3, Integration ×2); APT smoke skipped. mergeStateStatus is CLEAN.

Ready to merge.