fix(emulator): vectorize calcSpatialCov.matrix, removing O(n^3) redundant loop

[anshul23102]

Summary

Fixes #3964.

calcSpatialCov.matrix() contained a nested loop that was doing far more work than needed:

# Before
for (i in seq_len(nl)) {
  for (j in seq_len(nl)) {
    H[i, ] <- tau * exp(-psi * d[i, ])
  }
}

The inner j loop was completely redundant: on every iteration it overwrote H[i, ] with the same value, repeating the same assignment nl times. The outer i loop was also unnecessary since R applies arithmetic operations to entire matrices at once.

The fix replaces both loops with a single vectorized expression:

# After
tau * exp(-psi * d)

This produces identical output in O(n^2) instead of O(n^3). The already-vectorized calcSpatialCov.list sibling method was the model for this approach.

Regarding the second bug mentioned in #3964 (brittle unnamed run.id in do.call in sensitivity.R): both do.call calls at lines 309 and 470 already pass run.id = run.id as a named argument, so no change is needed there.

Changes

• modules/emulator/R/calcSpatialCov.matrix.R: replace nested loops with one vectorized expression
• modules/emulator/tests/testthat/test-calcSpatialCov.R: new test file covering correctness, output dimensions, diagonal values, symmetry, and agreement with calcSpatialCov.list
• modules/emulator/NEWS.md: changelog entry

Test plan

• New tests in test-calcSpatialCov.R pass
• Output of calcSpatialCov.matrix() matches calcSpatialCov.list() for a single-component distance matrix

[infotroph]

Is this a duplicate of #3969?

[anshul23102]

Yes, this is a duplicate of #3969 - I missed that PR when I opened this. Closing in favor of that one.