Need interpretable error if `LagrangianParticles` are initialized with invalid positions

[glwagner]

Right now if we do this, we get an obscure error from calc_correct_velocity_u (which is already a confusing name --- what's the "incorrect" velocity?)

nested task error: BoundsError: attempt to access 608-element OffsetArray(::StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64}, -3:604) with eltype Float64 with indices -3:604 at index [-58]
    Stacktrace:
      [1] throw_boundserror(A::OffsetArrays.OffsetVector{Float64, StepRangeLen{Float64, Base.TwicePrecision{Float64}, Base.TwicePrecision{Float64}, Int64}}, I::Tuple{Int64})
        @ Base ./abstractarray.jl:691
      [2] checkbounds
        @ ./abstractarray.jl:656 [inlined]
      [3] getindex
        @ ~/.julia/packages/OffsetArrays/80Lkc/src/OffsetArrays.jl:435 [inlined]
      [4] calc_correct_velocity_u
        @ ~/.julia/packages/Oceananigans/HYqGf/src/LagrangianParticleTracking/update_particle_properties.jl:104 [inlined]
      [5] update_particle_position!
        @ ~/.julia/packages/Oceananigans/HYqGf/src/LagrangianParticleTracking/update_particle_properties.jl:72 [inlined]

This probably has to be checked in model constructors, since LagrangianParticles doesn't know about grid.

[simone-silvestri]

Future work on the particles module is also to write some tests for immersed boundary particle tracking

[glwagner]

We should also decide whether to automatically convert particle arrays to CuArray for GPU architectures. I think probably yes.