fix: write Terraform plan artifacts to the run's unique temp dir

[jcoffi]

Fixes #202

Type of Change

• Bug Fix
• New Feature
• Refactor
• Documentation

What this changes

tf_initplan() in modules/tfwrapper.py wrote its plan artifacts (tfplan.bin, tfplan.json, tfgraph.dot) into os.path.dirname(temp_dir.name). temp_dir is the per-process tempfile.TemporaryDirectory, so taking its parent collapses the path back to the shared temp root (usually /tmp). The three artifacts therefore landed at fixed, shared paths.

Run two terravision processes at the same time and they write to the same /tmp/tfplan.bin (and .json/.dot), so they overwrite each other's plan and graph partway through a run. In practice this shows up as corrupted or mismatched diagrams when terravision runs concurrently, for example across parallel CI jobs sharing one runner.

The fix uses temp_dir.name directly, so each run keeps its artifacts inside its own unique directory.

-        tempdir = os.path.dirname(temp_dir.name)
+        tempdir = temp_dir.name

Why this is safe

• temp_dir.name is already the directory terravision stores as tfdata["terraform_init_dir"] (same file, line 371), so the plan artifacts now sit alongside the init data they belong with.
• _decode_plan() reads the same three local path variables, so it follows the new location automatically.
• read_tfsource() locates modules via terraform_init_dir, not these artifact paths, so module resolution is unaffected.
• No public interface or output changes. Behaviour for a single run is identical.

Testing

• python -m black --check modules/tfwrapper.py passes.
• python -m py_compile modules/tfwrapper.py passes.
• No automated regression test added. The bug lives inside tf_initplan(), where the artifact paths are local variables built inline between real terraform subprocess calls, so there is no clean seam to assert on without a small refactor. This PR is kept to the minimal one-line fix on purpose. I am happy to follow up with a refactor that extracts the path construction into a helper plus a regression test if you would prefer that.

A sibling bug left untouched

modules/tgwrapper.py has the same fixed-path pattern for Terragrunt (/tmp/tg_tfplan.bin and friends, around line 682). I left it out to keep this PR to a single change, but it is worth a separate fix.

Checklist

All Submissions:

• Checked for other open PRs covering the same change
• Written Documentation/Tests (see the Testing note above)
• Done my own code review
• Disclosed AI tool/model use (below)

AI Assistance Declaration

• Tools used: opencode (CLI coding agent)
• Model: Claude Opus 4 (anthropic/claude-opus-4-8)
• Scope: at the user's request, traced the bug through tf_initplan / _decode_plan / read_tfsource, made the one-line fix, and wrote this PR.

Checklist for Changes to Core Features:

• Minor one-line bugfix, raised directly as the template allows for minor fixes
• PR is focused on a single change
• Explanation of what and why is included above
• New tests not added (rationale in Testing)
• Verified the single-run path by tracing every reader of the changed paths