Address sea-level review feedback in land-mask generation

[Copilot]

This updates the sea-level handling introduced in the PR to match the linked review thread. It restores backward-compatible NULL behavior in make_land_mask(), documents supported sea_level inputs more clearly, and hardens get_sea_level() around invalid time ranges and malformed baseline data.

• make_land_mask() compatibility and docs

  • Treat sea_level = NULL as Spratt2016 to preserve prior caller behavior.
  • Clarify that sea_level may be:
    • "Spratt2016"
    • "Clark2025"
    • a numeric vector with the same length as time_bp

• get_sea_level() validation

  • Reject future times (time_bp > 0) explicitly instead of relying on interpolation failure.
  • Keep the existing lower-bound checks while making them tolerant of NA values in the range test.

• Baseline rescaling safety

  • Require the internal sea-level dataset to contain exactly one time_bp == 0 row before rescaling to present-day sea level.
  • Fail fast on malformed baseline data instead of returning ambiguous results.

• Focused regression coverage

  • Add coverage for sea_level = NULL behaving as Spratt2016.
  • Add coverage for present vs future time_bp.
  • Add coverage for malformed baseline datasets.

Example:

# preserved backward-compatible behavior
mask <- make_land_mask(relief_rast, time_bp = c(-1000, -10000), sea_level = NULL)

# explicit validation for future times
get_sea_level(1000)
# Error: time_bp should be in the past