First, you're correct in saying that it's not linked to CSMA/CD.
Second, you referenced a common, but incorrect belief that CSMA/CD was the reason for 10Base-T [half-duplex] 100m limit. This was a reason for - as you called it - classic Ethernet network length of 2500m (with ample margin - minimum frame of 64 bytes at 10Mb/s would 'occupy' around 11000m of cable - or to word it differently - collision would be heard back by sender at about middle of transmission)1.
So why 100m? It is linked to the electrical interface and signal characteristic outlined in the standard. One of the ideas behind twisted pair was to use existing cabling - and 100m was around max length that still satisfied parameters like attenuation, crosstalk, etc.
14.4.1 Overview
The medium for 10BASE-T is twisted-pair wiring. A significant number of 10BASE-T networks are installed utilizing in-place unshielded telephone wiring and typical telephony installation practices, the end to- end path including different types of wiring, cable connectors, and cross connects must be considered.(...omitted)
14.4.2 Transmission parameters
Each simplex link segment shall have the following characteristics. All characteristics specified apply to the total simplex link segment unless otherwise noted. These characteristics are generally met by 100 m of twisted-pair cable composed of 0.5 mm [24 AWG] twisted pairs.
That probably got carried over to newer/related (like EIA/TIA mentioned) standards (although I have no hard proof of that).
I also found an interesting section in the Ethernet/IEEE 802.2 Family AMD Handbook confirming that 100m was not set in stone:
AM79C940 10Base-T interface
(...omitted) when Low Receive Threshold bit is set, (...) sensitivity of the the 10Base-T MAU receiver is increased. This allows longer line lengths to be employed, exceeding the 100m target distance of normal 10Base-T (assuming typical 24AWG cable)
1 Of course propagation delay had its role in twisted pair too, hence 5-4-3 rule used in hub only networks.