Disallow mixing offset and non-offset axes in conv input

[martinholters]

In reference to #583, but doesn't fix it. Rather, it explicitly throws an error for any input where we're not quite sure what it should do. That would leave us the opportunity to make a more informed choice later when (and if) concrete use cases for mixed offset/non-offset inputs are reported.

[codecov]

Codecov Report

Attention: Patch coverage is 93.33333% with 1 line in your changes missing coverage. Please review.

Project coverage is 97.84%. Comparing base (a98e64f) to head (4c19923).
Report is 9 commits behind head on master.

Files with missing lines	Patch %	Lines
src/dspbase.jl	92.85%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #586   +/-   ##
=======================================
  Coverage   97.84%   97.84%           
=======================================
  Files          19       19           
  Lines        3249     3254    +5     
=======================================
+ Hits         3179     3184    +5     
  Misses         70       70           

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

---

🚨 Try these New Features:

• Flaky Tests Detection - Detect and resolve failed and flaky tests

[martinholters]

As fill(true) doesn't have any axes, this should arguably work. However, conv currently first brings all arguments to the same dimensionality, turning fill(true) into [true] here, and the axis compatibility check then says no. Fixing that is probably not important and certainly beyond the scope of this PR, but I thought leaving a @test_broken as a reminder might be in order.

[wheeheee]

LGTM, could also @test_throws for the ArgumentError

[martinholters]

LGTM, could also @test_throws for the ArgumentError

For what currently is the @test_broken, you mean? Yeah, I'm not sure whether we'd like that to work or not. This way it's maybe clearer that we're still undecided.

[wheeheee]

I was referring to conv_axis_with_offset(a::Any)

[martinholters]

Ah, ok. Well, could of course directly test conv_axis_with_offset, like

@test_throws ArgumentError conv_axis_with_offset("foo")

But I don't think that adds much value. And for any axis that might actually appear in the wild, we might conceivably support it in the future, so not much point in explicitly testing that it errors now.

[wheeheee]

Yeah that makes sense too. Merge?