Give element blocks one canonical order, and match physics by name

[lxmota]

The bug

Per-block mesh data lives in Dicts keyed by block name, while the block names themselves live in an ordered Vector. Dict iteration is hash order, so values(mesh.element_conns) and mesh.element_block_names are two different orderings of the same blocks — and the code that indexes blocks positionally was mixing them.

On a three-block mesh with blocks b1, b2, b3 holding 1, 2 and 3 elements:

block_names(fspace)          = ["b1", "b2", "b3"]     # file order
[num_elements(fspace, b)...] = [3, 2, 1]              # Dict order

Block b's name and block b's elements were different blocks. Three consequences:

1. FunctionSpace took elem_conns and elem_id_maps from values(...) but block names and ref_fes from element_block_names. With mixed element types across blocks this pairs a block's connectivity with another block's reference element.
2. BCBookKeeping / _setup_sideset numbered blocks with enumerate(values(mesh.element_id_maps)) and then used that number to index element_block_names, so a side set resolved to the wrong block name — and from there the wrong connectivity and reference element for its tractions.
3. write_to_file wrote block names in file order but the blocks themselves in Dict order, pairing each block with another block's name on round-trip.

Two blocks is the boundary case where it happens not to bite: every multi-block fixture in the suite is two blocks named block_1/block_2, for which hash order equals file order. That is why 51k passing tests never saw it.

The fix

block_names(mesh) is now the single source of truth for what "block b" means, with block_conns(mesh) / block_id_maps(mesh) returning per-block data in that same order. Every positional use is routed through them, and block_names(fspace) exposes the same order downstream.

With one order established, physics and properties can be matched to blocks by name rather than by position. Parameters now permutes a supplied NamedTuple into block order and requires its keys to be exactly the block names. This resolves the standing # TODO re-arrange physics tuple to match fspaces when appropriate — previously a NamedTuple written in any other order silently gave each block another block's material, and nothing downstream could flag it, since a run with swapped materials converges perfectly well to the wrong answer.

Tests

test/TestBlockOrdering.jl builds a three-block fixture (via Exodus, at test time — no new binary in the repo) whose hash order and file order differ, with deliberately different element counts per block so a permutation shows up as a size and not just a name. It covers the mesh/function-space invariant, the NamedTuple permutation, and each rejection case.

Verified failing on main before the fix:

Expression: [num_elements(V, b) for b = 1:length(BLOCK_NAMES)] == BLOCK_SIZES
 Evaluated: [3, 2, 1] == [1, 2, 3]

Full suite green after: 51824 passed.

Behavior changes for downstream users

• p.physics / p.properties are keyed by real block names instead of region_N placeholders. (region_N appeared nowhere else in this repo and nothing indexed by it, but it was observable.)
• A physics/properties NamedTuple whose keys are not exactly the block names now raises BlockMismatchError instead of being accepted.
• An unnamed Tuple is rejected rather than assumed to be in block order.

Version bumped to 0.15.0 on that account rather than a patch release.

Not addressed

The p_degree > 1 path in FunctionSpace still builds connectivity from values(create_higher_order_mesh(...)). That path is untested and already broken independently (create_higher_order_mesh destructures a Vector{String} as if it were an id => name pair, and keys a Dict{Symbol,...} with Strings), so straightening out only its ordering would have implied a fix it does not have.

[codecov]

Codecov Report

❌ Patch coverage is 87.09677% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 69.07%. Comparing base (157760c) to head (fd58593).

Files with missing lines	Patch %	Lines
src/Parameters.jl	81.48%	5 Missing ⚠️
src/FunctionSpaces.jl	84.61%	2 Missing ⚠️
src/meshes/AMRMesh.jl	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #321      +/-   ##
==========================================
+ Coverage   68.99%   69.07%   +0.07%     
==========================================
  Files          55       55              
  Lines        6112     6134      +22     
==========================================
+ Hits         4217     4237      +20     
- Misses       1895     1897       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[cmhamel]

Thanks for shoring up this buggy part of the package @lxmota!

No worries about the stuff you didn't address. That's dead code anyway and will be removed once I finalize #319 as an initial stab towards mixed space fem