Migrate internal dat$run$nw references to use get_network()/set_network() accessors

[smjenness]

Summary

EpiModel provides get_network() and set_network() S3 methods for netsim_dat objects (defined in R/get.R), which correctly abstract over the tergmLite vs. full-mode network storage. However, 31 occurrences across 9 internal R files still access dat$run$nw directly, bypassing these accessors.

Motivation

• Consistency: Extension module authors read internal code to learn patterns. Direct dat$run$nw access teaches the wrong pattern and may break if the internal structure changes.
• Abstraction: The accessors handle tergmLite/full-mode differences internally. Direct access requires each call site to handle this distinction, which is error-prone.
• Maintainability: If the internal storage structure changes, only the accessor implementations need updating rather than 31 call sites.

Scope

Files with direct dat$run$nw access (31 occurrences total):

File	Count
R/net.mod.init.R	7
R/saveout.R	4
R/update.R	4
R/net.fn.utils.R	3
R/net.mod.nwupdate.R	2
R/edgelists.R	1

Also 6 occurrences in inst/JSS/ and tests/testthat/ (lower priority).

Similarly, dat$run$el is accessed directly in several files and could benefit from the same treatment, though get_edgelist() already exists for read access.

Notes

• The vignettes have already been updated to use get_network()/set_network() exclusively (PR Updates to Package Vignettes #976).
• Some internal call sites (e.g., net.mod.init.R initializing dat$run$nw <- nws) may need a dedicated initialization path rather than set_network(), since the object may not be fully constructed yet. These should be evaluated case by case.

[AdrienLeGuillou]

I started to implement these changes in #978 but I realize they are not completely trivial due to the difference between networks and networksLite. The latter requires ancillary data like net_attr.

I think it could be useful to have a more general update where we list every way in which EpiModel interact with network and then implement generic method for each case.

I'll try to look into that idea and see if it make actual sense.

[smjenness]

I guess on further reflection -> don't feel like you need to be absolute and strict on this change. If there are some instances that don't use get/set_network that's fine. The main change is in the most visible user-facing modules such as those we might most readily modify and feature in the gallery, SISMID materials, and package vignettes (the last being how I initially flagged this as a problem to solve).

[AdrienLeGuillou]

noted, I'll do it with that in mind

[smjenness]

Closed with #978