fix(mix): scope dep library names relative to the project

[sh41]

Fixes the absolute path reported in #3926.

A Mix dep's project library was named <dep> [<absolute content root URL>]. That name is also the file name the platform derives for .idea/libraries, and what every <orderEntry> repeats, so none of it survived a different checkout path. $PROJECT_DIR$ cannot help: path macros only collapse a path at the start of an attribute value, and the file name is derived before serialization. The scope is now relative to the project base directory — the same anchor $PROJECT_DIR$ resolves to.

Names written by older versions are replaced on the next sync. Only the <dep> [<token>] shape this plugin writes is ever touched; a bare name may be a user's own library and is left alone.

Also here, because the rename is not usable without them:

• Tools > Elixir > Rebuild Mix Libraries, since nothing but a deps/ change or first-time project setup could rebuild libraries, leaving a project with wrong ones no way back.
• Repair on project open, escalating to a full sync only when libraries and content roots disagree, so a healthy project does no dependency scanning. It waits for the JPS model first — the IDE applies the real .iml over the cached model seconds after open, and would otherwise discard the sync.

Verified against two real WSL projects: a single-root one migrated 43 old-scheme entries and persisted without a manual save; a two-root umbrella with an out-of-tree sibling module produced . and ../<sibling> scopes.

This is enough for the reporter to test. Remaining clean-up — order entries left by deps removed from mix.exs, and orphaned libraries scoped outside every content root — is deferred to #3927.

[github-actions]

Test Results

Commit 4d7b338f · 12 legs

✅ Required legs: 0 failing tests, all 12 ran to completion.

	count
tests per leg	7,077 - 7,181
runs	85,318
✅ passing runs	85,318
❌ failing runs	0
💤 skipped runs	0
❌ distinct tests failing somewhere	0
result files	5,184
wall clock (longest leg)	15m 55s
runner time (all legs)	2h 40m

runs counts executions across the whole matrix, so a test present in every leg is one test and as many runs as there are legs - that is the figure to read for a matrix whose job is per-version validation. distinct tests failing somewhere collapses it back: a test that fails on four legs counts once. Every row spans required and informational legs alike; only the required-leg line above decides whether this commit can merge.

Required legs

leg	os	IDEA	Elixir	OTP	status	tests	✅	❌	wall clock
1.13.4+25.3.2.21	ubuntu-22.04	2025.3.6	1.13.4	25.3.2.21	✅ passed	7,077	7,077	0	12m 23s
1.14.5+26.2.5.21	ubuntu-22.04	2025.3.6	1.14.5	26.2.5.21	✅ passed	7,100	7,100	0	14m 21s
1.15.8+26.2.5.21	ubuntu-22.04	2025.3.6	1.15.8	26.2.5.21	✅ passed	7,103	7,103	0	15m 55s
1.16.3+26.2.5.21	ubuntu-22.04	2025.3.6	1.16.3	26.2.5.21	✅ passed	7,105	7,105	0	11m 59s
1.17.3+27.1.2	ubuntu-22.04	2025.3.6	1.17.3	27.1.2	✅ passed	7,119	7,119	0	13m 18s
1.18.4+27.3.4	ubuntu-22.04	2025.3.6	1.18.4	27.3.4	✅ passed	7,138	7,138	0	12m 25s
1.19.5+28.4	ubuntu-22.04	2025.3.6	1.19.5	28.4	✅ passed	7,175	7,175	0	12m 22s
1.20.2+29.0	ubuntu-22.04	2025.3.6	1.20.2	29.0	✅ passed	7,181	7,181	0	12m 55s
IDEA 2025.3.6	ubuntu-22.04	2025.3.6	1.13.4	24.3.4.6	✅ passed	7,080	7,080	0	12m 18s
IDEA 2026.1.4	ubuntu-22.04	2026.1.4	1.13.4	24.3.4.6	✅ passed	7,080	7,080	0	12m 11s
IDEA 2026.2	ubuntu-22.04	2026.2	1.13.4	24.3.4.6	✅ passed	7,080	7,080	0	14m 24s
Win25, IDEA 2025.3.6	windows-2025	2025.3.6	1.13.4	24.3.4.6	✅ passed	7,080	7,080	0	15m 35s

Counts only - for which tests failed, open a leg’s Test Results (…) check, or the Failed tests summary on its test (…) job.