fix: preserve source entity binding in forward fallback

[ZetabS]

Problem

Refs #620.

The documented platform-specific hmPlatforms example can fail when a
user-scoped complex forward includes a fromClass whose class key never appears
in the user's aspect tree and is not explicitly registered in den.classes.

For example, a user may include hmPlatforms and define only hmLinux content.
The generated forward for another class, such as hmDarwin, then has no
collected source modules when there is no hmDarwin key in that user's aspect
tree and hmDarwin is not registered as a class.

In that case getCollectedSource returns [], so resolveSourceFallback
rewalks the source aspect from the parent scope to keep fleet visibility. That
fallback used bindings = { }, dropping the source scope's entity binding. For
a user-scoped forward, losing the user binding can make the spawn collapse
into a self-parent edge:

den: spawnNode spawn root equals its parent scope 'host=igloo,system=x86_64-linux'
— a self-parent edge collapses policyBoundAncestor to null and yields zero fleet peers.
The seed ctx likely lost its child binding (e.g. `user`).

Fix

Restore the source scope's entity binding during complex forward fallback
rewalk.

The binding reconstruction mirrors the existing
mkDrained drained-spawn path:

den/nix/lib/aspects/fx/resolve.nix

Lines 677 to 679 in 70dfa3a

	sctx = scopeContexts.${scopeId} or { };
	ownKind = scopeEntityKind.${scopeId} or null;
	ownRecord = if ownKind == null then null else sctx.${ownKind} or null;

den/nix/lib/aspects/fx/resolve.nix

Lines 702 to 704 in 70dfa3a

	bindings = {
	${ownKind} = ownRecord;
	};

The fallback now follows the same pattern: look up the source scope's entity
kind, read that entity from scopeContexts, and pass it as the spawn binding.

This keeps the existing parent-scope spawn behavior for fleet visibility, while
preserving bindings like user when the sourceScopeId is a child entity
scope.

Tests

Added deadbugs.hm-platform-forward covering:

• the docs hmPlatforms shape: active hmLinux forwards, inactive hmDarwin
  stays inert
• an undeclared/unused source class forwarded into homeManager, which should
  be inert rather than crash

Both new regression tests fail on the test-only commit with spawnNode spawn root equals its parent scope and pass with this fix.

Validated locally:

• just ci deadbugs.hm-platform-forward.test-doc-example-forwards-linux-only
• just ci deadbugs.hm-platform-forward.test-unknown-source-class-to-home-manager-is-inert
• just fmt
• just ci: 1036/1036 successful

AI assistance

I used AI assistance while preparing this contribution:

• ChatGPT 5.5 Thinking helped me understand the complex forward fallback path,
  analyze the dropped source scope entity binding, and notice the existing
  mkDrained binding reconstruction pattern.
• Codex GPT-5.5 medium helped with contribution workflow, test strategy,
  fail/pass comparison, test naming/comments, and English wording.

I also used AI to translate and polish English because I am not fluent in
written English. I reviewed the final text, understand the code changes and PR
description, removed repetition, kept the diff minimal and focused, and ran the
final tests locally.

I Gyeongmin Song, explicitly acknowledge the following:

• My contribution is AI generated and I mention which models/mcp-servers/tools were used.
• I recognize AI generated contribution as my own and am myself legally accountable for it.
• My contribution aligns with Den LICENSE and does not violate any other projects licenses (e.g. GPL)
• I will truthfully answer and provide any requested details on how contribution was generated.

[ZetabS]

Sorry, I forgot to add the ai-assisted tag.

[sini]

This looks solid enough to me.

[sini]

I'll merge it once CI passes. Thank you for your contribution! :)