odb: more refactoring of tmg_conn

[AcKoucher]

Summary

General Organization

• Make all tmg_conn method definitions live in tmg_conn.cpp.
• Only write the definition in .h if it is a oneliner.
• Leave public only the truly public methods.

Structural Changes

• Decouple the target net setter from the top-level API. We now have tmg_conn::setNet.
• Make static void function tmg_getDriveTerm a method of tmg_conn.
• Remove all tmg_conn methods that just forwarded graph APIs. They were somewhat hiding that the DFS is referring to the ConnectionGraph.
• Split tmg_conn::loadNet into:

tmg_conn::clear()
tmg_conn::loadWire()
tmg_conn::loadTerminals()

• tmg_conn::clear() now resets the entire state i.e., all member variables and it's called at the end of the top-level function.

Removals

• Remove unused checker tmg_conn::checkConnOrdered. Not only it wasn't being used but the name oversold what it did. It basically just checked if a wire path's first point was a terminal. If we want a proper checker for normalized connectivity, we can write it ourselves. This method was defined alone in tmg_conn_w.cpp which I removed.
• TestOrderWires no longer tries to exercise the post-ordering path that was supposed to test the method cited above.
• tmg_conn::isConnected was also dead.
• Redundant net state setters in top-level function.

Type of Change

• Refactoring

Impact

None.

Verification

• I have verified that the local build succeeds (./etc/Build.sh).
• I have run the relevant tests and they pass.
• My code follows the repository's formatting guidelines.
• I have signed my commits (DCO).

Related Issues

None.

[gemini-code-assist]

Code Review

This pull request refactors the tmg_conn class by removing tmg_conn_w.cpp and restructuring how nets are analyzed. Specifically, it separates net assignment (setNet) from analysis (analyzeNet), moves several graph-related traversal methods directly into ConnectionGraph, and cleans up unused methods. The review feedback highlights a critical issue where clear() does not reset net-specific state variables (like connected_ and has_special_wires_), leading to potential state leakage across reused instances. Additionally, several defensive null checks are recommended for connection_graph_ to prevent potential null pointer dereferences in checkVisited(), getDisconnectedStart(), and copyWireIdToVisitedShorts().

[maliberty]

Codex:

• src/odb/src/db/tmg_conn.cpp:39: bug: When loadWire() produces no points, analyzeNet() now calls clear() without resetting the net’s disconnected flag. Previously this path set setDisconnected(false). dbWire::destroy() clears the ordered flag but leaves the disconnected flag intact, so ordering a formerly disconnected net after its wire is removed leaves it incorrectly marked disconnected. Please restore the empty wire state update and add a regression test.

[AcKoucher]

Fixed. As discussed, the bug was in the wire destroyer.

[maliberty]

Are you killing off disconnected? If so you should remove the APIs and storage too.

[AcKoucher]

I didn't realized that it was dead.

[AcKoucher]

@maliberty Killing off disconnected would imply more dead code in orderWires that I'll have to also remove here and the PR would get overloaded. I reverted the last commit and did as Codex suggested. I'll remove the flag in a subsequent hopefully removal-only PR.